You’ll hear many things about artificial intelligence (AI), depending on who you choose to listen to.
The doomsayers will tell you that AI is mankind’s last invention. They see a dystopian world where superintelligence will run riot and either eliminate us or enslave us. The less discerning envisage a future filled with job-killing robots with everyday functions taken over without a need for humans.
In this welter of confusion comes a book that paints a picture of AI that is far less complicated and not as threatening. “Prediction Machines” envisions a far better future compared to where we are today.
The three authors – Ajay Agrawal, Joshua Gans, and Avi Goldfarb – are academicians at Rotman School of Management, each has a background in economics. Together they also run Creative Destruction Lab (CDL), a Toronto-based accelerator that boasts of having the greatest concentration of AI startups on the planet.
Now in case you are wondering why to read a book on AI written by economists, then let me clear the air for you. Where others see crazy fads and radical transformations, economists see what has changed. So the AI might lead to pathbreaking turnarounds in business, but, the thing that matters to economists is what cost benefits will accrue to companies.
The key thing to understand is that even with recent leaps in the fields of machine intelligence and deep learning, we are still quite far off from machines achieving the human-level intelligence. A stage often
But instead of riding the fear-mongering bandwagon about the disastrous impact of the AI, the authors focus on its positive implications in business.
They note that what we call AI is at its heart a prediction technology. The output of machine learning is
Prediction is a key component of intelligence and that’s the reason why technologists prefer to call it AI.
You might not have realized it but predictions are everywhere. Every day we make numerous decisions on the back of predictions.
Here’s how the trio defines it:
“Prediction is the process of filling in missing information. Prediction takes information you have, called data, and uses it to generate information you don’t have.”Prediction Machines
Predictions provide foundational inputs to decision-making and economists provide frameworks for decision-making.
Cheaper Data, More Predictions
The costs of prediction are dropping. Why? Because the main input – data – is getting cheaper. More quality data means better predictions.
It’s not hard to figure out that when predictions are cheap, there will be more and more of them. The availability of massive data enables AI experts to convert many non-prediction problems into prediction problems.
Take, for example, autonomous driving. It is not a new phenomenon. Leading auto manufacturers have used autonomous vehicles inside the plants for quite some time. But they have not been tested outside those controlled environments until recently.
Car manufacturers like Tesla transformed self-driving into a prediction problem. With the help of sensors, cameras and GPS, the Tesla AI was fed massive data to learn and improve consequentially.
Another interesting example where the manifestation of cheap predictions will be prominent, according to authors, is Amazon. Amazon is no stranger to AI. The company was one of the first to use technology to drive its product recommendations.
If you order stuff from Amazon, you’d have noticed ‘related to items you’ve viewed‘ header on the home page. That’s Amazon’s prediction technology at work.
In the future, however, authors expect Amazon to ship you stuff based on the frequency of your past purchases before you even log on to shop. Shipping before shopping, you know. It sounds implausible today, but at the pace Amazon is building its AI, it will become a possibility in the near future.
Predictions will improve Human Judgments
Machines don’t make judgments, humans do.
The optimal scenario will see both humans and machines working together with humans in supervisory roles.
Authors supply the example of credit card frauds to make their point. Machines can flag a dubious transaction basis the past data – frequency, timings and quantum of transactions by a consumer. But how do you separate a fraudulent transaction which has a 90% chance of being correct from the one with 10% chance? You need a human in the loop.
Further, such situations involve a clear trade-off scenario for the company. If you settle for blocking what is indeed an unauthorized transaction, then you save yourselves the pain of recovering costs. However, if you are wrong, you have on your hands a dissatisfied and angry customer who is ready to sever his ties with the company.
To this effect, authors state, “As prediction machines make predictions increasingly better, faster, and cheaper, the value of human judgment will increase because we will need more of it.
A Future brimming with Trade-offs?
The book leaves you with a disconcerting thought that the future will involve a lot of trade-offs. It’s quite possible that policy-makers or other interventionists might settle for insalubrious choices at times. For example, wealth accumulation vs wealth distribution.
AI or Prediction Machines are skill-biased technologies. This has deep implications as AI tools might disproportionately boost the productivity of a select few and leave several others by the wayside. Further, it’s no longer a conjecture now that AI will take over certain tasks from humans. For the remaining jobs, the authors predict there will be a scramble.
If these scenarios play out, we might witness a manifold increase in the capital of tech-owners and capitalists. At the same time, those on the fringes will get further marginalized as their incomes will get hit.
Unlike Max Tegmark’s bestseller “Life 3.0” which focused on vast realms of AI like Artificial General Intelligence and Superintelligence, “Prediction Machines” is about the narrow AI, the kind which is being used today.
A general audience will indeed find Prediction Machines much of interest. However, Data scientists or Machine learning practitioners may not glean a great deal of new material in this book.
Furthermore, as economists, the authors are indeed adept at uncovering potential unintended consequences of prediction technology and this makes for interesting reading.
Authors also avoid using the term AI as such in excess. Since in its existing avatar, most AI tools help accomplish specific tasks only – that of solidifying predictions – they settle for “Prediction Machines”.