The Times Australia
The Times World News

.

Philosophers have studied 'counterfactuals' for decades. Will they help us unlock the mysteries of AI?

  • Written by Sam Baron, Associate Professor, Philosophy of Science, Australian Catholic University
Philosophers have studied 'counterfactuals' for decades. Will they help us unlock the mysteries of AI?

Artificial intelligence is increasingly being rolled out all around the world to help make decisions in our lives, whether it’s loan decisions by banks[1], medical diagnoses[2], or US law enforcement predicting a criminal’s likelihood of re-offending[3].

Yet many AI systems are black boxes: no one understands how they work. This has led to a demand for “explainable AI”, so we can understand why an AI model yielded a specific output, and what biases may have played a role.

Explainable AI is a growing branch of AI research. But what’s perhaps less well known is the role philosophy plays in its development.

Specifically, one idea called “counterfactual explanation” is often put forth as a solution to the black box problems. But once you understand the philosophy behind it, you can start to understand why it falls short.

Why explanations matter

When AI is used to make life-changing decisions, the people impacted deserve an explanation of how that decision was reached. This was recently recognised through the European Union’s General Data Protection Regulation[4], which supports an individual’s right to explanation.

The need for explanation was also highlighted in the Robodebt case in Australia[5], where an algorithm was used to predict debt levels for individuals receiving social security. The system made many mistakes, placing people into debt who shouldn’t have been.

It was only once the algorithm was fully explained that the mistake was identified – but by then the damage had been done. The outcome was so damaging it led to a royal commission[6] being established in August 2022.

In the Robodebt case, the algorithm in question was fairly straightforward and could be explained. We should not expect this to always be the case going forward. Current AI models using machine-learning to process data are much more sophisticated.

Read more: Not everything we call AI is actually 'artificial intelligence'. Here's what you need to know[7]

The big, glaring black box

Suppose a person named Sara applies for a loan. The bank asks her to provide information including her marital status, debt level, income, savings, home address and age.

The bank then feeds this information into an AI system, which returns a credit score. The score is low and is used to disqualify Sara for the loan, but neither Sara nor the bank employees know why the system scored Sara so low.

Unlike with Robodebt, the algorithm being used here may be extremely complicated and not easily explained. There is therefore no straightforward way to know whether it has made a mistake, and Sara has no way to get the information she needs to argue against the decision.

This scenario isn’t entirely hypothetical: loan decisions are likely to be outsourced to algorithms in the US, and there’s a real risk they will encode bias[8]. To mitigate risk, we must try to explain how they work.

Read more: Everyone's having a field day with ChatGPT – but nobody knows how it actually works[9]

The counterfactual approach

Broadly speaking, there are two types of approaches[10] to explainable AI. One involves cracking open a system and studying its internal components to discern how it works. But this usually isn’t possible due to the sheer complexity of many AI systems.

The other approach is to leave the system unopened, and instead study its inputs and outputs, looking for patterns. The “counterfactual” method falls under this approach.

Counterfactuals are claims about what would happen if things had played out differently. In an AI context, this means considering how the output from an AI system might be different if it receives different inputs. We can then supposedly use this to explain why the system produced the result it did.

One example of a counterfactual would be to ask what the world might be like had the internet never been developed. Shutterstock

Suppose the bank feeds its AI system different (manipulated) information about Sara. From this, the bank works out the smallest change Sara would need to get a positive outcome would be to increase her income.

The bank can then apparently use this as an explanation: Sara’s loan was denied because her income was too low. Had her income been higher, she would have been granted a loan.

Such counterfactual explanations[11] are being seriously considered[12] as a way of satisfying the demand for explainable AI, including in cases of loan applications and using AI to make scientific discoveries[13].

However, as researchers have argued, the counterfactual approach is inadequate[14].

Correlation and explanation

When we consider changes to the inputs of an AI system and how they translate into outputs, we manage to gather information about correlations. But, as the old adage goes, correlation is not causation.

The reason that’s a problem is because work in philosophy suggests causation is tightly connected to explanation[15]. To explain why an event occurred, we need to know what caused it.

On this basis, it may be a mistake for the bank to tell Sara her loan was denied because her income was too low. All it can really say with confidence is that income and credit score are correlated – and Sara is still left without an explanation for her poor result.

What’s needed is a way to turn information about counterfactuals and correlations into explanatory information.

The future of explainable AI

With time we can expect AI to be used more for hiring decisions, visa applications, promotions and state and federal funding decisions, among other things.

A lack of explanation for these decisions threatens to substantially increase the injustice people will experience. After all, without explanations we can’t correct mistakes made when using AI. Fortunately, philosophy can help.

Explanation has been a central topic of philosophical study[16] over the last century. Philosophers have designed a range of methods for extracting explanatory information from a sea of correlations, and have developed sophisticated theories about how explanation works.

A great deal of this work has focused on the relationship between counterfactuals and explanation. I’ve developed work on this[17] myself. By drawing on philosophical insights, we may be able to develop better approaches to explainable AI.

At present, however, there’s not enough overlap between philosophy and computer science on this topic. If we want to tackle injustice head-on, we’ll need a more integrated approach that combines work in these fields.

Read more: When self-driving cars crash, who's responsible? Courts and insurers need to know what's inside the 'black box'[18]

References

  1. ^ loan decisions by banks (www.afr.com)
  2. ^ medical diagnoses (www.healthcareoutlook.net)
  3. ^ criminal’s likelihood of re-offending (www.technologyreview.com)
  4. ^ General Data Protection Regulation (gdpr.eu)
  5. ^ Robodebt case in Australia (theconversation.com)
  6. ^ royal commission (www.abc.net.au)
  7. ^ Not everything we call AI is actually 'artificial intelligence'. Here's what you need to know (theconversation.com)
  8. ^ they will encode bias (hbr.org)
  9. ^ Everyone's having a field day with ChatGPT – but nobody knows how it actually works (theconversation.com)
  10. ^ two types of approaches (news.mit.edu)
  11. ^ counterfactual explanations (jolt.law.harvard.edu)
  12. ^ seriously considered (www.abc.net.au)
  13. ^ scientific discoveries (www.chemistryworld.com)
  14. ^ inadequate (arxiv.org)
  15. ^ tightly connected to explanation (www.jstor.org)
  16. ^ topic of philosophical study (plato.stanford.edu)
  17. ^ work on this (quod.lib.umich.edu)
  18. ^ When self-driving cars crash, who's responsible? Courts and insurers need to know what's inside the 'black box' (theconversation.com)

Read more https://theconversation.com/philosophers-have-studied-counterfactuals-for-decades-will-they-help-us-unlock-the-mysteries-of-ai-196392

Times Magazine

September Sunset Polo

International Polo Tour To Bridge Historic Sport, Life-Changing Philanthropy, and Breath-Taking Beauty On Saturday, September 6th, history will be made as the International Polo Tour (IPT), a sports leader headquartered here in South Florida...

5 Ways Microsoft Fabric Simplifies Your Data Analytics Workflow

In today's data-driven world, businesses are constantly seeking ways to streamline their data analytics processes. The sheer volume and complexity of data can be overwhelming, often leading to bottlenecks and inefficiencies. Enter the innovative da...

7 Questions to Ask Before You Sign IT Support Companies in Sydney

Choosing an IT partner can feel like buying an insurance policy you hope you never need. The right choice keeps your team productive, your data safe, and your budget predictable. The wrong choice shows up as slow tickets, surprise bills, and risky sh...

Choosing the Right Legal Aid Lawyer in Sutherland Shire: Key Considerations

Legal aid services play an essential role in ensuring access to justice for all. For people in the Sutherland Shire who may not have the financial means to pay for private legal assistance, legal aid ensures that everyone has access to representa...

Watercolor vs. Oil vs. Digital: Which Medium Fits Your Pet's Personality?

When it comes to immortalizing your pet’s unique personality in art, choosing the right medium is essential. Each artistic medium, whether watercolor, oil, or digital, has distinct qualities that can bring out the spirit of your furry friend in dif...

DIY Is In: How Aussie Parents Are Redefining Birthday Parties

When planning his daughter’s birthday, Rich opted for a DIY approach, inspired by her love for drawing maps and giving clues. Their weekend tradition of hiding treats at home sparked the idea, and with a pirate ship playground already chosen as t...

The Times Features

Do you really need a dental check-up and clean every 6 months?

Just over half of Australian adults[1] saw a dental practitioner in the past 12 months, most commonly for a check-up[2]. But have you been told you should get a check-up and c...

What is a Compounding Pharmacy and Why Do You Need One in Melbourne?

Ever picked up a prescription and thought, this pill is too big, too bitter, or full of things I cannot have? That is where a compounding chemist becomes important. A compounding p...

Deep Cleaning vs Regular Cleaning: Which One Do Perth Homes Really Need?

Whether you live in a coastal home in Cottesloe or a modern apartment in East Perth, keeping your living space clean isn’t just about aesthetics, it’s essential for your health and...

Rubber vs Concrete Wheel Stops: Which is Better for Your Car Park?

When it comes to setting up a car park in Perth, wheel stops are a small feature that make a big difference. From improving driver accuracy to preventing costly damage, the right c...

Not all processed foods are bad for you. Here’s what you can tell from reading the label

If you follow wellness content on social media or in the news, you’ve probably heard that processed food is not just unhealthy, but can cause serious harm. Eating a diet domin...

What happens if I eat too much protein?

The hype around protein[1] intake doesn’t seem to be going away. Social media is full of people urging you to eat more protein, including via supplements such as protein sha...