Artificial Intelligence: Boom or Doom?
Professor Amit Joshi looks at the hype and reality of AI at Orchestrating Winning Performance
Can AI think? Does it imitate the human brain? Will it take decision making away from us?
Don’t worry, we’re not there yet. We currently have no such thing as general artificial intelligence and we won’t for the next few decades.
Self-driving cars are cool but they’re no terminator. It is true that computers can recognize faces better, play chess better, and maybe eventually drive a car better than human beings. But this narrow artificial intelligence that exists already today will not rise up and take control from human beings.
How smart is AI?
Today’s AI algorithms are not even as smart as chimps yet; it can only focus on specific tasks.
As Roy Amara, an American researcher, famously said: “We tend to overestimate the effects of technology in the short run but underestimate them in the long run.”
What I un-famously say is that we tend to overestimate what AI is capable of but underestimate how it can be applied. Even though general AI is just a fantasy, we completely underestimate the applications of our current AI.
Even today’s dumb AI already has the power to change the world.
What does AI actually do?
By and large, applied AI is a super human pattern recognition machine.
Much of what humans do on a regular basis is pattern recognition and prediction. Sales and hiring are examples of jobs that include a lot of these.
With self driving vehicles, we have reduced something as complicated as driving a car down to prediction. AI does not judge, it can only predict.
Anything that classifies as prediction, we should let AI do because it will do it better than us. Anything that is judgement, humans should do because AI can’t do it and won’t be able to for the predictable future. AI will make us better because it will enable us to spend more time on things we’re actually good at.
Applicable AI has arrived and is here to stay: Think recommendations for songs and shows on your favorite apps. We will have tools for applicable AI similar to excel in the coming 3-5 years.
Will AI take your job?
From a technology perspective, AI could replace some of the most common jobs on the planet like cashiers, waiters, store clerks, heavy truck drivers, and low-level lawyers.
To ensure employment in the future, our entire education system today needs to be rethought. We have to move to a system of lifelong learning to stay ahead of how fast technology is moving.
Our emphasis needs to be placed on qualities such as critical thinking, communications and empathy. A huge set of jobs in the future will focused on human to human contact.
What about privacy? Privacy is dead. Ethics is definitely also an issue with AI. Questions like ‘Should I really put all of my customers’ health records into an AI system?’ will arise more and more. Privacy is one thing. Ethics is another side of the same coin. There are guidelines. They will evolve. We don’t have a full picture of what AI will be applied for, so we don’t know exactly what needs to be regulated yet.
What do we need to do right away?
- Make AI more transparent: Now it is a black box. It is not explainable. We tell it what to do but not how to do it. And if you try to find out how it comes to its conclusions, it can’t tell you. A work around is to use black box for relatively low risk tasks until we can make it more transparent.
- Do our best to eliminate bias in AI. Humans create AI and transfer their bias. AI can also learn bias on its own. Creating unbiased AI is extremely hard. The first step is to be extremely careful with training data.
- Figure out how to make AI choose the least bad option. AI will be faced with dilemmas on a daily basis. Should a self-driving car turn right to avoid an accident and injure or kill two people or turn left and do the same for three? What about choosing between elderly or children? Choices like these will arise and how do humans program for this? We have to ensure that there are ample feedback loops in the AI so that humans can intervene quickly and often as required.
Bottom line: with great power comes great responsibility. And even if AI never improves, we already have tremendous capabilities.