Category: Machine Learning

A machine wrote this post – OpenAi’s ChatGPT

Blog post by OpenAI’s ChatGPT
Introduction by Jo Brodie

ChatGPT image by Alexandra_Koch from Pixaba

1. Jo wrote this…

ChatGPT is an AI chatbot which can have a conversation with you using everyday (natural) language. It has access to the huge amounts of digital information found on the internet and its ability to use language to write good answers in response to questions has been tested and improved by giving it feedback. It learns, and refines its output.

You don’t need to programme it, you can just ask it a question (giving it a prompt) and, after a moment’s ‘thinking’ it will produce an answer. Lots of people have been testing ChatGPT by asking it questions that we already know the answer to, to see how well it can do. It’s actually quite impressive! If you think it’s made a mistake or misunderstood your prompt you can reword your question or give it more information to help it improve its answer.

Can ChatGPT write an article that’s suitable for our magazine or website?

I gave ChatGPT the following prompt: “Please write a 200 to 500 word article suitable for the CS4FN magazine” and moments later it generated a completely new article about machine learning. That told me it knew (or at least could find out) that CS4FN has something to do with computer science. I’m not surprised that it picked machine learning as a topic given that it is a machine that can learn. I didn’t tell it what to write about, it came up with that idea itself.

The article it gave me contained some technical words and not all of them were explained, so I asked it to rewrite the article for a younger audience (who might not have come across some of the words yet).

The result is below – what do you think?

2. ChatGPT wrote this…

From Logic to Learning: The Incredible Advancements of Artificial Intelligence

Artificial Intelligence (AI) has come a long way since it was first invented in the 1950s. At the beginning, AI was based on logic and rules. Experts in a specific field would give the computer a set of instructions to help it make decisions. But, these rules were limited and couldn’t handle complex situations.

Today, AI has seen a huge shift towards machine learning. This means that instead of following rules, AI learns from data. This shift happened because computers got stronger, and more data became available. The breakthroughs in machine learning algorithms helped too.

Machine learning has two categories: supervised learning and unsupervised learning. In supervised learning, the computer learns from labeled data. This means that the correct answer is given to the computer for each input. Once the computer has learned enough, it can predict the answer for new data. Supervised learning is useful for applications like image and speech recognition.

In unsupervised learning, the computer learns from unlabeled data. This means that there is no correct answer provided. The goal of unsupervised learning is to find patterns in the data. This type of learning is useful in applications like anomaly detection and clustering.

Deep learning is a special type of machine learning that uses neural networks. Neural networks are inspired by the human brain and have layers of connected nodes. Each node applies a mathematical function to its inputs and sends the result to the next layer of nodes. Deep learning has revolutionized AI, leading to breakthroughs in areas like natural language processing, computer vision, and game playing.

Even though machine learning has achieved so much, there are still challenges to overcome. One challenge is the lack of transparency in machine learning models. Unlike rule-based systems, it can be hard to understand how a machine learning model made a decision. This can be a problem, especially in areas like healthcare where the decision-making process needs to be trustworthy and explainable.

Another challenge is the need for a lot of data to train machine learning models. While data is easy to find in some areas, like social media, it can be hard to come by in other areas like medical imaging. This lack of data can make machine learning less effective in some applications.

In conclusion, AI has come a long way since the 1950s. We’ve gone from rule-based systems to machine learning, which has led to breakthroughs in many areas. Although there are still challenges to overcome, the future of AI is exciting!

The first attempt is below…

The Evolution of Artificial Intelligence: From Rule-Based to Machine Learning

Artificial Intelligence (AI) has come a long way since its inception in the 1950s. Initially, AI was developed based on rule-based systems, where experts in a specific domain would provide rules to the system to help it make decisions. While these systems worked well in some areas, they were limited in their ability to handle complex problems and adapt to new situations.

In recent years, AI has seen a significant shift towards machine learning, where systems learn from data rather than relying on pre-defined rules. This shift has been driven by advances in computer hardware, the availability of large amounts of data, and breakthroughs in machine learning algorithms.

Machine learning has two main categories: supervised learning and unsupervised learning. In supervised learning, the system is trained on labeled data, where the correct answer is provided for each input. The system uses this labeled data to learn a function that maps inputs to outputs. Once trained, the system can make predictions on new, unseen data. This approach has been successful in a wide range of applications, such as image and speech recognition.

In unsupervised learning, the system is trained on unlabeled data, where no correct answer is provided. The goal of unsupervised learning is to discover patterns and structure in the data. This approach has been useful in areas such as anomaly detection and clustering.

Deep learning is a subset of machine learning that uses neural networks to learn from data. Neural networks are inspired by the structure of the human brain and consist of layers of interconnected nodes. Each node applies a mathematical function to its inputs and passes the result to the next layer of nodes. Deep learning has revolutionized the field of AI, enabling breakthroughs in areas such as natural language processing, computer vision, and game playing.

Despite the many successes of machine learning, there are still limitations and challenges to overcome. One challenge is the lack of interpretability of machine learning models. Unlike rule-based systems, it can be difficult to understand how a machine learning model arrived at a particular decision. This lack of transparency can be problematic in areas such as healthcare, where the decision-making process needs to be explainable and trustworthy.

Another challenge is the need for large amounts of data to train machine learning models. While data is abundant in some domains, such as social media, it can be scarce in others, such as medical imaging. This scarcity of data can limit the effectiveness of machine learning approaches in certain applications.

In conclusion, the evolution of AI from rule-based systems to machine learning has enabled breakthroughs in many areas and holds tremendous potential for the future. However, there are still challenges to overcome, particularly in the areas of interpretability and data scarcity. Addressing these challenges will be critical to realizing the full potential of AI in the years to come.

Further reading

Chatbot or Cheatbot?
Meet the chatbots

This blog is funded through EPSRC grant EP/W033615/1.

What’s that bird? Ask your phone – birdsong-recognition apps


by Dan Stowell, Queen Mary University of London

Could your smartphone automatically tell you what species of bird is singing outside your window? If so how?

Mobile phones contain microphones to pick up your voice. That means they should be able to pick up the sound of birds singing too, right? And maybe even decide which bird is which?

Smartphone apps exist that promise to do just this. They record a sound, analyse it, and tell you which species of bird they think it is most likely to be. But a smartphone doesn’t have the sophisticated brain that we have, evolved over millions of years to understand the world around us. A smartphone has to be programmed by someone to do everything it does. So if you had to program an app to recognise bird sounds, how would you do it? There are two very different ways computer scientists have devised to do this kind of decision making and they are used by researchers for all sorts of applications from diagnosing medical problems to recognising suspicious behaviour in CCTV images. Both ways are used by phone apps to recognise bird song that you can already buy.

Robin image by Darren Coleshill from Pixabay
The sound of the European robin (Erithacus rubecula) better known as robin redbreast, from Wikipedia.

Write down all the rules

If you ask a birdwatcher how to identify a blackbird’s sound, they will tell you specific rules. “It’s high-pitched, not low-pitched.” “It lasts a few seconds and then there’s a silent gap before it does it again.” “It’s twittery and complex, not just a single note.” So if we wrote down all those rules in a recipe for the machine to follow, each rule a little program that could say “Yes, I’m true for that sound”, an app combining them could decide when a sound matches all the rules and when it doesn’t.

Young blackbird in Oxfordshire, from Wikipedia
The sound of a European blackbird (Turdus merula) singing merrily in Finland, from Wikipedia (song 1).

This is called an ‘expert system’ approach. One difficulty is that it can take a lot of time and effort to actually write down enough rules for enough birds: there are hundreds of bird species in the UK alone! Each would need lots of rules to be hand crafted. It also needs lots of input from bird experts to get the rules exactly right. Even then it’s not always possible for people to put into words what makes a sound special. Could you write down exactly what makes you recognise your friends’ voices, and what makes them different from everyone else’s? Probably not! However, this approach can be good because you know exactly what reasons the computer is using when it makes decisions.

This is very different from the other approach which is…

Show it lots of examples

A lot of modern systems use the idea of ‘machine learning’, which means that instead of writing rules down, we create a system that can somehow ‘learn’ what the correct answer should be. We just give it lots of different examples to learn from, telling it what each one is. Once it has seen enough examples to get it right often enough, we let it loose on things we don’t know in advance. This approach is inspired by how the brain works. We know that brains are good at learning, so why not do what they do!

One difficulty with this is that you can’t always be sure how the machine comes up with its decisions. Often the software is a ‘black box’ that gives you an answer but doesn’t tell you what justifies that answer. Is it really listening to the same aspects of the sound as we do? How would we know?

On the other hand, perhaps that’s the great thing about this approach: a computer might be able to give you the right answer without you having to tell it exactly how to do that!

It means we don’t need to write down a ‘recipe’ for every sound we want to detect. If it can learn from examples, and get the answer right when it hears new examples, isn’t that all we need?

Which way is best?

There are hundreds of bird species that you might hear in the UK alone, and many more in tropical countries. Human experts take many years to learn which sound means which bird. It’s a difficult thing to do!

So which approach should your smartphone use if you want it to help identify birds around you? You can find phone apps that use one approach or another. It’s very hard to measure exactly which approach is best, because the conditions change so much. Which one works best when there’s noisy traffic in the background? Which one works best when lots of birds sing together? Which one works best if the bird is singing in a different ‘dialect’ from the examples we used when we created the system?

One way to answer the question is to provide phone apps to people and to see which apps they find most useful. So companies and researchers are creating apps using the ways they hope will work best. The market may well then make the decision. How would you decide?

This article was originally published on the CS4FN website and can also be found on pages 10 and 11 of Issue 21 of the CS4FN magazine ‘Computing sounds wild’. You can download a free PDF copy of the magazine (below), or any of our other free material at our downloads site.

Further bird- (& computing-) themed reading
🐦🐤🦜🦉

What are birds actually saying?
CS4FN Advent – Day 7 – Computing for the birds: dawn chorus, birds as data carriers and a Google April Fool (plus a puzzle!)

Related Magazine …

This blog is funded through EPSRC grant EP/W033615/1.

Recognising (and addressing) bias in facial recognition tech – the Gender Shades Audit #BlackHistoryMonth

The five shades used for skin tone emojis

Some people have a neurological condition called face blindness (also known as ‘prosopagnosia’) which means that they are unable to recognise people, even those they know well – this can include their own face in the mirror! They only know who someone is once they start to speak but until then they can’t be sure who it is. They can certainly detect faces though, but they might struggle to classify them in terms of gender or ethnicity. In general though, most people actually have an exceptionally good ability to detect and recognise faces, so good in fact that we even detect faces when they’re not actually there – this is called pareidolia – perhaps you see a surprised face in this picture of USB sockets below.

A unit containing four sockets, 2 USB and 2 for a microphone and speakers.
Happy, though surprised, sockets

What if facial recognition technology isn’t as good at recognising faces as it has sometimes been claimed to be? If the technology is being used in the criminal justice system, and gets the identification wrong, this can cause serious problems for people (see Robert Williams’ story in “Facing up to the problems of recognising faces“).

In 2018 Joy Buolamwini and Timnit Gebru shared the results of research they’d done, testing three different commercial facial recognition systems. They found that these systems were much more likely to wrongly classify darker-skinned female faces compared to lighter- or darker-skinned male faces. In other words, the systems were not reliable.

“The findings raise questions about how today’s neural networks, which … (look for) patterns in huge data sets, are trained and evaluated.”

Study finds gender and skin-type bias in commercial artificial-intelligence systems
(11 February 2018) MIT News

The Gender Shades Audit

Facial recognition systems are trained to detect, classify and even recognise faces using a bank of photographs of people. Joy and Timnit examined two banks of images used to train facial recognition systems and found that around 80 per cent of the photos used were of people with lighter coloured skin. 

If the photographs aren’t fairly balanced in terms of having a range of people of different gender and ethnicity then the resulting technologies will inherit that bias too. Effectively the systems here were being trained to recognise light-skinned people.

The Pilot Parliaments Benchmark

They decided to create their own set of images and wanted to ensure that these covered a wide range of skin tones and had an equal mix of men and women (‘gender parity’). They did this by selecting photographs of members of various parliaments around the world which are known to have a reasonably equal mix of men and women, and selected parliaments from countries with predominantly darker skinned people (Rwanda, Senegal and South Africa) and from countries with predominantly lighter-skinned people (Iceland, Finland and Sweden). 

They labelled all the photos according to gender (they did have to make some assumptions based on name and appearance if pronouns weren’t available) and used the Fitzpatrick scale (see Different shades, below) to classify skin tones. The result was a set of photographs labelled as dark male, dark female, light male, light female with a roughly equal mix across all four categories – this time, 53 per cent of the people were light-skinned (male and female).

A composite image showing the range of skin tone classifications with the Fitzpatrick scale on top and the skin tone emojis below.

Different shades

The Fitzpatrick skin tone scale (top) is used by dermatologists (skin specialists) as a way of classifying how someone’s skin responds to ultraviolet light. There are six points on the scale with 1 being the lightest skin and 6 being the darkest. People whose skin tone has a lower Fitzpatrick score are more likely to burn in the sun and not tan, and are also at greater risk of melanoma (skin cancer). People with higher scores have darker skin which is less likely to burn and they have a lower risk of skin cancer. 

Below it is a variation of the Fitzpatrick scale, with five points, which is used to create the skin tone emojis that you’ll find on most messaging apps in addition to the ‘default’ yellow. 

Testing three face recognition systems

Joy and Timnit tested the three commercial face recognition systems against their new database of photographs – a fair test of a wide range of faces that a recognition system might come across – and this is where they found that the systems were less able to correctly identify particular groups of people. The systems were very good at spotting lighter-skinned men, and darker skinned men, but were less able to correctly identify darker-skinned women, and women overall.  

These tools, trained on sets of data that had a bias built into them, inherited those biases and this affected how well they worked. Joy and Timnit published the results of their research and it was picked up and discussed in the news as people began to realise the extent of the problem, and what this might mean for the ways in which facial recognition tech is used. 

“An audit of commercial facial-analysis tools found that dark-skinned faces are misclassified at a much higher rate than are faces from any other group. Four years on, the study is shaping research, regulation and commercial practices.”

The unseen Black faces of AI algorithms (19 October 2022) Nature

There is some good news though. The three companies made changes to improve their facial recognition technology systems and several US cities have already banned the use of this tech in criminal investigations, and more cities are calling for it too. People around the world are becoming more aware of the limitations of this type of technology and the harms to which it may be (perhaps unintentionally) put and are calling for better regulation of these systems.

Further reading

• Study finds gender and skin-type bias in commercial artificial-intelligence systems (11 February 2018) MIT News
• Facial recognition software is biased towards white men, researcher finds (11 February 2018) The Verge
• Go read this special Nature issue on racism in science (21 October 2022) The Verge

More technical articles

• Joy Buolamwini and Timnit Gebru (2018) Gender Shades: Intersectional Accuracy Disparities in Commercial Gender Classification, Proceedings of Machine Learning Research 81:1-15.
• The unseen Black faces of AI algorithms (19 October 2022) Nature News & Views

See more in ‘Celebrating Diversity in Computing‘

We have free posters to download and some information about the different people who’ve helped make modern computing what it is today.

Screenshot showing the vibrant blue posters on the left and the muted sepia-toned posters on the right

Or click here: Celebrating diversity in computing

This blog is funded through EPSRC grant EP/W033615/1.

The machines can translate now

by Paul Curzon, Queen Mary University of London

(From the cs4fn archive)

“The Machines can translate now…
…I SAID ‘THE MACHINES CAN TRANSLATE NOW'”

Portion of the Rosetta Stone which has the same text written in three languages.

The stereotype of the Englishman abroad when confronted by someone who doesn’t speak English is just to say it louder. That could soon be a thing of the past as portable devices start to gain speech recognition skills and as the machines get better at translating between languages.

Traditionally machine translation has involved professional human linguists manually writing lots of translation rules for the machines to follow. Recently there have been great advances in what is known as statistical machine translation where the machine learns the translations rules automatically. It does this using a parallel corpus*: just lots of pairs of sentences; one a sentence in the original language, the other its translation. Parallel corpora* are extracted from multi-lingual news sources like the BBC web site where professional human translators have done the translations.

Let’s look at an example translation of the accompanying original arabic:

Machine Translation: Baghdad 1-1 (AFP) – The official Iraqi news agency reported that the Chinese vice-president of the Revolutionary Command Council in Iraq, Izzat Ibrahim, met today in Baghdad, chairman of the Saudi Export Development Center, Abdel Rahman al-Zamil.

Human Translation: Baghdad 1-1 (AFP) – Iraq’s official news agency reported that the Deputy Chairman of the Iraqi Revolutionary Command Council, Izzet Ibrahim, today met with Abdul Rahman al-Zamil, Managing Director of the Saudi Center for Export Development.

This example shows a sentence from an Arabic newspaper then its translation by the Queen Mary University of London’s statistical machine translator, and finally a translation by a professional human translator. The statistical translation does allow a reader to get a rough understanding of the original Arabic sentence. There are several mistakes, though.

Mistranslating the “Managing Director” of the export development center as its “chairman” is perhaps not too much of a problem. Mistranslating “Deputy Chairman” as the “Chinese vice-president” is very bad. That kind of mistranslation could easily lead to grave insults!

That reminds me of the point in ‘The Hitch-Hiker’s Guide to the Galaxy’ where Arthur Dent’s words “I seem to be having tremendous difficulty with my lifestyle,” slip down a wormhole in space-time to be overheard by the Vl’hurg commander across a conference table. Unfortunately this was understood in the Vl’hurg tongue as the most dreadful insult imaginable, resulting in them waging terrible war for centuries…

For now the human’s are still the best translators but the machines are learning from them fast!

*corpus and corpora = singular and plural for the word used to describe a collection of written texts, literally a ‘body’ of text. A corpus might be all the works written by one author, corpora might be works of several authors.

More on …

The Emoji Crystal Ball

Fairground fortune tellers claim to be able to tell a lot about you by staring into a crystal ball. They could tell far more about you (that wasn’t made up) by staring at your public social media profile. Even your use of emojis alone gives away something of who you are.

Reflective ball with dots of lights
Image by Hier und jetzt endet leider meine Reise auf Pixabay  from Pixabay

Walid Magdy’s research team at Edinburgh University are interested in how much people unknowingly give away about themselves when they use social media. They have found that it’s possible to work out an awful lot about you from your social media activity. One of their experiments involved exploring emojis. About a fifth of posts on Twitter include emojis, so they wondered if anything could be predicted about people, ignoring what they wrote and just looking at the emojis they used in their tweets. They found that the way people use emojis in twitter posts alone gives away whether they are male or female and their ethnic background.

They started by taking a large number of tweets known to be written by either men or women and stripped out the words, leaving only the emojis they used. They then counted how often each group used different emojis. The differences in use of each emoji seemed to be revealing as there was clearly a different pattern of use overall of each emoji by men and by women. Men and women each use some emojis much more than others.

Next, they used emoji data for some of the people to train a machine learning system (creating what is known as a classifier). The classifier was given all the emojis used by a person and told which were by men and which by women. It built up a detailed pattern of what a man’s emoji profile was like and similarly what a woman’s was like. 

Given a new set of tweets from a single person the classifier could then try to predict man or woman based on whether that profile was closer to the male pattern or closer to the female pattern of emoji use. Walid’s team found their emoji classifier’s predictions were right about 80% of the time – essentially it was as accurate as doing a similar thing based on the words they wrote. When they tried a similar experiment with ethnicity (was the person black, white or of another ethnicity) the predictions were even more accurate getting it right 84% of the time.

A lot can be worked out about you from apparently innocuous information that is publicly available as a result of your social media use. Even emojis give away something of who you are 😦

– Paul Curzon, Queen Mary University of London, Spring 2021

– Based on a talk given by Walid Magdy at QMUL, May 2021.

Is your healthcare algorithm racist?

Algorithms are taking over decision making, and this is especially so in healthcare. But could the algorithms be making biased decisions? Could their decisions be racist? Yes, and such algorithms are already being used.

A medical operation showing an anaesthetist and the head of the patient
Image by David Mark from Pixabay

There is now big money to be made from healthcare software. One of the biggest areas is in intelligent algorithms that help healthcare workers make decisions. Some even completely take over the decision making. In the US, software is used widely, for example, to predict who will most benefit from interventions. The more you help a patient the more it costs. Some people may just get better without extra help, but for others it means the difference between a disability that might have been avoided or not, or even life and death. How do you tell? It matters as money is limited, so someone has to choose. You need to be able to predict outcomes with or without potential treatments. That is the kind of thing that machine learning technology is generally good at. By looking at the history of lots and lots of past patients, their treatments and what happened, these artificial intelligence programs can spot the patterns in the data and then make predictions about new patients.

This is what current commercial software does. Ziad Obermeyer, from UC Berkeley, decided to investigate how well the systems made those decisions. Working with a team combining academics and clinicians, they looked specifically at the differences between black and white patients in one widely used system. It made decisions about whether to put patients on more expensive treatment programmes. What they found was that the system had a big racial bias in the decisions it made. For patients that were equally ill, it was much more likely to recommend white patients for treatment programmes.

One of the problems with machine learning approaches is it is hard to see why they make the decisions they do. They just look for patterns in data, and who knows what patterns they find to base their decisions on? The team had access to the data of a vast number of patients the algorithms had made recommendations about, the decisions made about them and the outcomes. This meant they could evaluate whether patients were treated fairly.

The data given to the algorithm specifically excluded race, supposedly to stop it making decisions on colour of skin. However, despite not having that information, that was ultimately what it was doing. How?

The team found that its decision-making was based on predicting healthcare costs rather than how ill people actually were. The greater the cost saving of putting a person on a treatment programme, the more likely it was to recommend them. At first sight, this seems reasonable, given the aim is to make best use of a limited budget. The system was totally fair in allocating treatment based on cost. However, when the team looked at how ill people were, black people had to be much sicker before they would be recommended for help. There are lots of reasons more money might be spent on white people, so skewing the system. For example, they may be more likely to seek treatment earlier or more often. Being poor means it can be harder to seek healthcare due to difficulties getting to hospital, difficulties taking time off work, etc. If more black people in the data used to train the system are poor then this will lead to them seeking help less, so less is being spent on them. The system had spotted patterns like this and that was how it was making decisions. Even though it wasn’t told who was black and white, it had learnt to be biased.

There is an easy way to fix the system. Instead of including data about costs and having it use that as the basis of decision making, you can use direct measures of how ill a person is: for example, using the number of different conditions the patient is suffering from and the rule of thumb that the more complications you have, the more you will benefit from treatment. The researchers showed that if the system was trained this way instead, the racial bias disappeared. Access to healthcare became much fairer.

If we are going to allow machines to take healthcare decisions for us based on their predictions, we have to make sure we know how they make those predictions, and make sure they are fair. You should not lose the chance of the help you need just because of your ethnicity, or because you are poor. We must take care not to build racist algorithms. Just because computers aren’t human doesn’t mean they can’t be humane.

– Paul Curzon, Queen Mary University of London, Spring 2021

Download Issue 27 of the cs4fn magazine on Smart Health here.

This post and issue 27 of the cs4fn magazine have been funded by EPSRC as part of the PAMBAYESIAN project.

I’m feeling Moo-dy today

It has long been an aim of computer scientists to develop software that can work out how a person is feeling. Are you happy or sad, frustrated or lonely? If the software can tell then it can adapt to you moods, changing its behaviour or offering advice. Suresh Neethirajan from Wageningen University in the Netherlands has gone step further. He has developed a program that detects the emotions of farm animals.

Image by Couleur from Pixabay 

Working out how someone is feeling is called “Sentiment Analysis” and there are lots of ways computer scientists have tried to do it. One way is based on looking at the words people speak or write. The way people speak, such as the tone of voice also gives information about emotions. Another way is based on our facial expressions and body language. A simple version of sentiment analysis involves working out whether someone is feeling a positive emotion (like being happy or excited) versus a negative emotions (such as being sad or angry) rather than trying to determine the precise emotion.

Applications range from deciding how a person might vote to predicting what they might buy. A more futuristic use is to help medics make healthcare decisions. When the patient says they are aren’t feeling too bad, are they actually fine or are they just being stoical, for example? And how much pain or stress are they actually suffering?

But why would you want to know the emotions of animals? One really important application is to know when an animal is, or is not, in distress. Knowing that can help a farmer look after that animal better, but also work out the best way to better look after animals more generally. It might help farmers design nicer living conditions, but also work out more humane ways to slaughter animals that involves the least suffering. Avoiding cruel conditions is reason on its own, but with happy farm animals you might also improve the yield of milk, quality of meat or how many offspring animals have in their lifetime. A farmer certainly shouldn’t want their animals to be so upset they start to self harm, which can be a problem when animals are kept in poor conditions. Not only is it cruel it can lead to infections which costs money to treat. It also spreads resistance to antibiotics. Having accurate ways to quickly and remotely detect how animals are feeling would be a big step forward for animal welfare.

But how to do it? While some scientists are actually working on understanding animal language, recognising body language is an easier first step to understand animal emotions. A lot is actually known about animal expressions and body language, and what they mean. If a dog is wagging its tail, then it is happy, for example. Suresh focussed on facial expressions in cows and pigs. What kind of expressions do they have? Cows, for example, are likely to be relaxed if their eyes are half-closed, and their ears are backwards or hung-down. If you can see the whites of their eyes, on the other hand then they are probably stressed. Pigs that are moving their ears around very quickly, by contrast, are likely to be stressed. If their ears are hanging and flipping in the direction of their eyes, though, then they are in a much more neutral state.

There are lots of steps to go through in creating a system to recognise emotions. The first for Suresh was to collect lots of pictures of cows and pigs from different farms. He collected almost 4000 images from farms in Canada, the USA and India. Each image was labelled by human experts according to whether it showed a positive, neutral and negative emotional state of the animal, based on what was already known about how animal expressions link to their emotions.

Sophisticated image processing software was then used to automatically pick out the animals’ faces as well as locate the individual features, such as eyes and ears. The orientation and other properties of those facial features, such as whether ears were hanging down or up is also determined. This processed data is then fed into a machine learning system to train it on this data. The fact that it was labelled meant the program knew what a human judged the different expressions to mean in terms of emotions, and so could then work out how patterns in the data that represented each animal state.

Once trained the system was then given new images without the labels to judge how accurate it was. It made a judgement and this was compared to the human judgement of the state. Human and machine agreed 86% of the time. More work is needed before such a system could be used on farms but it opens the possibility of using video cameras around a farm to raise the alarm when animals are suffering, for example.

Machine learning is helping humans in lots of ways. With systems like this machine learning could soon be helping animals live better lives too.

– Paul Curzon, Queen Mary University of London, Spring 2021

Standup Robots

‘How do robots eat pizza?’… ‘One byte at a time’. Computational Humour is real, but it’s not jokes about computers, it’s computers telling their own jokes.

Robot performing
Image from istockphoto

Computers can create art, stories, slogans and even magic tricks. But can computers perform themselves? Can robots invent their own jokes? Can they tell jokes?

Combining Artificial Intelligence, computational linguistics and humour studies (yes you can study how to be funny!) a team of Scottish researchers made an early attempt at computerised standup comedy! They came up with Standup (System to Augment Non Speakers Dialogue Using Puns): a program that generates riddles for kids with language difficulties. Standup has a dictionary and joke-building mechanism, but does not perform, it just creates the jokes. You will have to judge for yourself as to whether the puns are funny. You can download the software from here. What makes a pun funny? It is a about the word having two meanings at exactly the same time in a sentence. It is also about generating an expectation that you then break: a key idea about what is at the core of creativity too.

A research team at Virginia Tech in the US created a system that started to learn about funny pictures. Having defined a ‘funniness score’ they created a computational model for humorous scenes, and trained it to predict funniness, perhaps with an eye to spotting pics for social media posting, or not.

But are there funny robots out there? Yes! RoboThespian programmed by researchers at Queen Mary University of London, and Data, created by researchers at Carnegie Mellon University are both robots programmed to do stand-up comedy. Data has a bank of jokes and responds to audience reaction. His developers don’t actually know what he will do when he performs, as he is learning all the time. At his first public gig, he got the crowd laughing, but his timing was poor. You can see his performance online, in a TED Talk.

RoboThespian did a gig at the London Barbican alongside human comedians. The performance was a live experiment to understand whether the robot could ‘work the audience’ as well as a human comedian. They found that even relatively small changes in the timing of delivery make a big difference to audience response.

What have these all got in common? Artificial Intelligence, machine learning and studies to understand what humour actually is, are being combined to make something that is funny. Comedy is perhaps the pinnacle of creativity. It’s certainly not easy for a human to write even one joke, so think how hard it is distill that skill into algorithms and train a computer to create loads of them.

You have to laugh!

Watch RoboThespian [EXTERNAL]

– Jane Waite, Queen Mary University of London, Summer 2017

Download Issue 22 of the cs4fn magazine “Creative Computing” here

Lots more computing jokes on our Teaching London Computing site

Sabine Hauert: Swarm Engineer

by Paul Curzon, Queen Mary University of London

Based on a 2016 talk by Sabine Hauert at the Royal Society

Sabine Hauert is a swarm engineer. She is fascinated by the idea of making use of swarms of robots. Watch a flock of birds and you see that they have both complex and beautiful behaviours. It helps them avoid predators very effectively, for example, so much so that many animals behave in a similar way. Predators struggle to fix on any one bird in all the chaotic swirling. Sabine’s team at the University of Bristol are exploring how we can solve our own engineering problems: from providing communication networks in a disaster zone to helping treat cancer, all based on the behaviours of swarms of animals.

A murmuration - a flock of starlings

Sabine realised that flocks of birds have properties that are really interesting to an engineer. Their ability to scale is one. It is often easy to come up with solutions to problems that work in a small ‘toy’ system, but when you want to use it for real, the size of the problem defeats you. With a flock, birds just keep arriving, and the flock keeps working, getting bigger and bigger. It is common to see thousands of Starlings behaving like this – around Brighton Pier most winter evenings, for example. Flocks can even be of millions of birds all swooping and swirling together, never colliding, always staying as a flock. It is an engineering solution that scales up to massive problems. If you can build a system to work like a flock, you will have a similar ability to scale.

Flocks of birds are also very robust. If one bird falls out of the sky, perhaps because it is caught by a predator, the flock itself doesn’t fail, it continues as if nothing happened. Compare that to most systems humans create. Remove one component from a car engine and it’s likely that you won’t be going anywhere. This kind of robustness from failure is often really important.

Swarms are an example of emergent behaviour. If you look at just one bird you can’t tell how the flock works as a whole. In fact, each is just following very simple rules. Each bird just tracks the positions of a few nearest neighbours using that information to make simple decisions about how to move. That is enough for the whole complex behaviour of the flock to emerge. Despite all that fast and furious movement, the birds never crash into each other. Fascinated, Sabine started to explore how swarms of robots might be used to solve problems for people.

Her first idea was to create swarms of flying robots to work as a communications network, providing wi-fi coverage in places it would otherwise be hard to set up a network. This might be a good solution in a disaster area, for example, where there is no other infrastructure, but communication is vital. You want it to scale over the whole disaster area quickly and easily, and it has to be robust. She set about creating a system to achieve this.

The robots she designed were very simple, fixed wing, propellor-powered model planes. Each had a compass so it knew which direction it was pointing and was able to talk to those nearest using wi-fi signals. It could also tell who its nearest neighbours were. The trick was to work out how to design the behaviour of one bird so that appropriate swarming behaviour emerged. At any time each had to decide how much to turn to avoid crashing into another but to maintain the flock, and coverage. You could try to work out the best rules by hand. Instead, Sabine turned to machine learning.

“Throwing those flying robots

and seeing them flock

was truly magical”

The idea of machine learning is that instead of trying to devise algorithms that solve problems yourself, you write an algorithm for how to learn. The program then learns for itself by trial and error the best solution. Sabine created a simple first program for her robots that gave them fairly random behaviour. The machine learning program then used a process modelled on evolution to gradually improve. After all evolution worked for animals! The way this is done is that variations on the initial behaviour are trialled in simulators and only the most successful are kept. Further random changes are made to those and the new versions trialled again. This is continued over thousands of generations, each generation getting that little bit better at flocking until eventually a behaviour of individual robots results that leads to them swarming together.

Sabine has now moved on to to thinking about a situation where swarms of trillions of individuals are needed: nanomedicine. She wants to create nanobots that are each smaller than the width of a strand of hair and can be injected into cancer patients. Once inside the body they will search out and stick themselves to tumour cells. The tumour cells gobble them up, at which point they deliver drugs directly inside the rogue cell. How do you make them behave in a way that gives the best cancer treatment though? For example, how do you stop them all just sticking to the same outer cancer cells? One way might be to give them a simple swarm behaviour that allows them to go to different depths and only then switch on their stickiness, allowing them to destroy all the cancer cells. This is the sort of thing Sabine’s team are experimenting with.

Swarm engineering has all sorts of other practical applications, and while Sabine is leading the way, some time soon we may need lots more swarm engineers, able to design swarm systems to solve specific problems. Might that be you?

Explore swarm behaviour using the Oxford Turtle system [EXTERNAL] (click the play button top centre) to see how to run a flocking simulation as well as program your own swarms.

–

More on …

What’s on your mind?

Telepathy is the supposed Extra Sensory Perception ability to read someone else’s mind at a distance. Whilst humans do not have that ability, brain-computer interaction researchers at Stanford have just made the high tech version a virtual reality.

Image by Andrei Cássia from Pixabay

It has long been know that by using brain implants or electrodes on a person’s head it is possible to tell the difference between simple thoughts. Thinking about moving parts of the body gives particularly useful brain signals. Thinking about moving your right arm, generates different signals to thinking about moving your left leg, for example, even if you are paralysed so cannot actually move at all. Telling two different things apart is enough to communicate – it is the basis of binary and so how all computer-to-computer communication is done. This led to the idea of the brain-computer interface where people communicate with and control a computer with their mind alone.

Stanford researchers made a big step forward in 2017, when they demonstrated that paralysed people could move a cursor on a screen by thinking of moving their hands in the appropriate direction. This created a point and click interface – a mind mouse – for the paralysed. Impressively, the speed and accuracy was as good as for people using keyboard applications

Stanford researchers have now gone a step even further and used the same idea to turn mental handwriting into actual typing. The person just thinks of writing letters with an imagined pen on imagined paper, the brain-computer interface then picks up the thoughts of subtle movements and the computer converts them into actual letters. Again the speed and accuracy is as good as most people can type. The paralysed participant concerned could communicate 18 words a minute and made virtually no mistakes at all: when the system was combined with auto-correction software, as we now all can use to correct our typing mistakes, it got letters right 99% of the time.

The system has been made possible by advances in both neuroscience and computer science. Recognising the letters being mind-written involves distinguishing very subtle differences in patterns of neurons firing in the brain. Recognising patterns is however, exactly what Machine Learning algorithms do. They are trained on lots of data and pick out patterns of similar data. If told what letter the person was actually trying to communicate then they can link that letter to the pattern detected. Here each letter will not lead to exactly the same pattern of brain signals firing each time, but they will largely clump together,. Other letters will also group but with slightly different patterns of firings. Once trained, the system works by taking the pattern of brain signals just seen and matching it to the nearest clumping pattern. The computer then guesses that the nearest clumping is the letter being communicated. If the system is highly accurate, as this one was at 94% (before autocorrection), then it means the patterns of most letters are very distinct. A letter being mind-written rarely fell into a brain pattern gap, which would have meant that letter could as easily have been the pattern of one letter as the other.

So a computer based “telepathy” is possible. But don’t expect us all to be able to communicate by mind alone over the internet any time soon. The approach involves having implants surgically inserted into the brain: in this case two computer chips connecting to your brain via 100 electrodes. The operation is a massive risk to take, and while perhaps justifiable for someone with a problem as severe as total paralysis, it is less obvious it is a good idea for anyone else. However, this shows at least it is possible to communicate written messages by mind alone, and once developed further could make life far better for severely disabled people in the future.

Yet again science fiction is no longer fantasy, it is possible, just not in the way the science fiction writers perhaps originally imagined by the power of a person’s mind alone.

– Paul Curzon, Queen Mary University of London, Spring 2021.