""

Distinguished Speaker Series: Avi Goldfarb

Avi Goldfarb, Ellison Professor of Marketing, Rotman School of Management, University of Toronto, gave the Distinguished Speaker Seminar in May.

AI: what does the future hold for individuals and organisations?

There is no shortage of hype around Artificial Intelligence (AI), and no shortage of anxiety either. As Avi Goldfarb summed it up, there is a lot of confusion about what AI really is, and what impact it is likely to have. ‘If you read the press the optimistic point of view is we have C3P0: a machine that does everything a human does … except C3PO listens to humans when they give them directions,’ he said. ‘The pessimistic scenario is more along the lines of the Matrix or the Terminator where machines become highly intelligent, decide “who needs these humans anyway?” and take over the world.’

But Goldfarb was not expecting either C3PO or Skynet any time soon. His view was that AI right now is all about prediction, and about making prediction so good and so cheap that it will transform business and society.

When things are cheaper, we do more of them

To explain, he took the audience back two generations to the original computer revolution: ‘What does your computer do? … It does one thing and only one. Your computer does arithmetic -- that is it.’ As a result, arithmetic is much easier and much cheaper, and therefore we do more of it, finding ‘all sorts of applications for arithmetic that [we] might not have thought of before.’

The first application for machine computers was to take over the arithmetic that was being done by humans, in places such as accountancy firms. And no doubt people at the time worried that the machines were going to do them out of jobs. But as Goldfarb pointed out, ‘there are still plenty of accountants in the world; because it turned out that the people who were doing the arithmetic were also best positioned to use that arithmetic to help companies’ strategies with tax policy and other things.’

Then ‘we found all sorts of applications of arithmetic that aren’t obviously arithmetical problems: it turns out games are arithmetic, music is arithmetic, pictures are arithmetic. All these things are arithmetic problems once the cost of arithmetic falls enough.’

The same, he said, was going to happen with the ‘prediction’ offered by AI -- essentially, ‘taking information that you have [data] to fill in information that you don’t have.’ The cheaper this prediction becomes, the more we are going to find other uses for it.

The ‘classic prediction problem’, he said, was working out credit-worthiness or insurance risk, and indeed these two tasks are increasingly being accomplished by AI. We are now discovering a vast range of other problems that we realise are, at heart, predictions problems. These range from medical diagnoses to object classification and even autonomous driving.

The importance of human judgement

Why is prediction valuable? According to Goldfarb it is to do with its usefulness in decision-making. And decisions, of course – large and small – are being made all the time. But prediction is not itself decision-making: AI will only take us so far. There is an important role for humans in ‘judgement’ – that is, ‘knowing what to do with the prediction once you have it … [deciding] how much more valuable one decision is relative to the other.’

At the most trivial level this is akin to deciding what to do with a weather prediction. If there is a chance of rain, ‘should we take an umbrella? How much better off are we if we have to carry around an umbrella and it doesn’t rain versus not carrying an umbrella and we get wet?’ More seriously it can be about deciding where to direct potentially life-saving resources.

So what does this mean for individuals (and their jobs) and organisations?

Goldfarb was sanguine about the future for jobs. ‘Overwhelmingly, AI companies provide tools that help with particular tasks within the workflow,’ he said. So companies might ‘unpack’ the different steps involved in any task, however complex, and ‘let’s say they figure out they can automate step number 132. That is a prediction task, so they can create a prediction tool to help with that prediction task. So if your job involves step number 132 you will no longer be needed, but what they have found is now you might need more humans for step number 131 and step number 133 because you are doing that middle step so much better.’

Turning the prediction dial

Prediction itself he could imagine getting better and better. Amazon’s recommendations tool, for example, is a prediction tool, and ‘at some point it will be good enough that they don’t have to wait for you to order …they will just ship the item right to your door because they are pretty sure that is the thing you want.’

Another example lies in the hiring process, where there are arguments for using AI to screen CVs: they are less prone to bias and fatigue. Goldfarb suggested that at some point the predictions will be so good that no one will have to apply for jobs any more. ‘Employers will say, “you are the person I want to hire and I have a pretty good prediction that you are going to say yes, and so I am just going to hire you” … that whole painful process goes away as predictions get better.’ Airport lounges and primary care physicians were also in his sights.

The challenge is to understand how quickly the dial in turning in each industry, he said, especially if you are ‘in an industry where the change is happening already or is likely to change in a year or two. In this case you need to really think about what the prediction problems are in your organisation and what particular task this tool is useful for. What compromises do you make to your overall strategy which could be better with more effective prediction?’