Clinton Castro is taking philosophy to places it hasn’t gone before.

Artificial intelligence. Big data. Machine learning. Algorithms. These may sound more like buzzwords than topics devoted to serious philosophical thought, but Castro — an assistant professor of philosophy — knows they present serious moral and ethical dilemmas.

Computer algorithms are able to find patterns in large quantities of data, helping humans do something that’s often humanly impossible. This new decision-making technology holds great promise. And also, quite a few problems and concerns about the more consequential decisions affecting human lives — from credit scores to hiring practices to the criminal justice system.

Castro is laser focused on how these technologies could be contributing to injustices in the justice system, especially when it comes to bail, parole and prison sentencing. The question that he keeps asking and exploring in his work is whether these systems are unfair and biased, and how they can perhaps be made to be fair and unbiased.

Castro explores two Florida criminal cases in a paper published in Ergo. The first involved the attempted theft of a bicycle and scooter by two young suspects. The second case involved a man stealing power tools from Home Depot.

In both cases, a computer program — which is used in different jurisdictions throughout the country — predicted the likelihood of each suspect committing a future crime. The two who attempted to steal the bike and scooter were considered high risk. The man who stole the tools, and who also carried previous convictions for multiple armed robberies, was considered low risk.

These risk assessment scores are predictions about the future that inform decisions for the present. In this case, high risk scores equated to a higher bond amount.

But, the program was far from accurate.

The “low risk” man whole stole the tools went on to steal again, racking up 30 felony counts for burglary and grand theft. He was sent to prison. The “high risk” defendants completed probation and haven’t committed other crimes.

The program has a history of misidentifying Black defendants as high risk at nearly twice the rate of white defendants. The suspects who attempted to steal the bike and scooter are Black. The man who stole the tools is white.

This got Castro thinking more about the bias that has crept into the machines we rely on. He began to explore possible ways to detect unfairness and focused in on two measures of fairness — classification parity and calibration.

Classification parity means that a system should not be making more mistakes about one group compared to another. For instance, it shouldn’t be more inaccurate for black defendants.

Calibration, as the name suggests, means the ‘scale’ is balanced for everyone. Scores should mean the same thing for everyone. For example, defendants considered “high risk” should reoffend at roughly the same rate, regardless of race.

“Surprisingly, classification parity and calibration can’t both be satisfied all the time,” Castro said. “Sometimes we must choose one over the other. This raises a host of difficult philosophical questions about fairness.”

Castro knows correcting machine bias can’t — and won’t — happen overnight. But, the first step is thinking about the issue. As he says, people can better detect mistakes and pinpoint potential biases by slowing down and taking the time to really notice potential biases and achieve a better understanding of what it even means for a system to be biased or unfair. 

For a field like philosophy that is, indeed, ancient, it may seem impossible that anyone could possibly be saying anything “new.” But, Castro is saying something new — all the while wrestling with decades old questions of right and wrong.

“As philosophers, it’s our job to understand what’s going on. To bring to the surface all of the assumptions that we make about what matters and what’s important,” Castro said. “It’s never easy to figure out the right thing to do. It’s especially hard with newer technologies and systems, but the first step in doing something about the issues is giving them proper attention and thought.”

The post Big data and the rise of the philosopher appeared first on Artificial Intelligence.

Artificial intelligence is everywhere – although often the reality feels somewhat underwhelming compared to the potential for what it might become. AI has the power to change the world, but it’s a gradual process.

Machine learning is often used as a synonym for artificial intelligence, but it’s actually a different, albeit related discipline. While artificial intelligence refers to a computer program able to “think” for itself without programmed instructions, machine learning is one process by which a computer can learn its trade.

The philosophy behind machine learning is similar to how you or I learn about something: by experiencing it. The difference is that for now at least, machines are specialised in learning about one or two things at a time.

Let’s take a hypothetical photos app that allows you to search its content. Type in the word “cat” and it shows pictures of cats even though there’s no labelling of the files themselves. How does the app know what cats look like? In short, it’s picked it up through machine learning. That is to say that the computer program was fed thousands of pictures of cats, and it began to notice patterns of what a cat looks like. Just as you might begin to spot similarities (tail, whiskers, ears), so does the machine, until eventually, you can show it random pictures taken from the internet, and it will be able to tell you if there’s a cat in the scene or not. Teach the same AI to recognise buildings, dogs, pizzas and people, and you’ve got a program that feels a bit like witchcraft.

This kind of machine learning is happening all around you, and you’re feeding various companies training data all the time. It’s the way that Google can predict typos in search boxes, and how Netflix knows the kind of shows people like you might enjoy watching.

That’s undeniably useful, but the world-changing aspects of machine learning is where the AI can learn to spot things that humans can’t. Earlier this year, I interviewed Saffron Technology’s CEO, Gayle Sheppard. Of the many examples she gave of how her company’s AI is innovating in the airline, health and banking sectors, one example sticks in my mind. By training the artificial intelligence with echocardiograms from patients with restrictive cardiomyopathy and constrictive pericarditis, the AI was able to spot the difference 96% of the time in two months. Human eyes get it right between 50 and 75% of the time.

In this instance, it was artificial intelligence which made the eventual diagnosis, but it couldn’t be done without the training. Machine learning is just one of the ways in which scientists are ensuring that artificial intelligence is as intelligent as it can be.

The post What’s the difference between AI and machine learning? appeared first on Artificial Intelligence.