It is predicted that in the next 50 years we will have built a brain that is smarter than us in every way. This will probably be the last thing we ever create and nobody knows what will happen. It keeps me awake at night and is probably the largest existential threat to humanity, leading AI expert Daniel Hulme told delegates at TechSmart 2019.
Daniel, who is CEO of Satalia and also runs a masters programme at UCL (University College London), views the world of artificial intelligence from both an academic and commercial viewpoint.
“There are two definitions that people often have of AI – the first one is popular and weak: getting computers to do things as good as or better than humans do. The second which is much better, is that AI is goal directed adaptive behaviour that works towards an objective. Good AI learns from its decisions both good and bad, and adapts its behaviour accordingly.”
Often, Daniel told delegates, companies don’t have technology or machine learning problems, they have decision-making problems and many are beginning to realise they have people working for them who have the wrong skills or the culture of the organisation does not motivate and inspire.
“We need organisations with a strong purpose which are empowering people for the future. It is not just about the technology – attracting the right talent is the most important thing. The challenge is how to attract this talent. You need the right culture to enable talent to thrive; traditional hierarchies are not conducive to innovation and the faster you can adapt the better.”
Data is only useful of you have it all in one place, analyse it, and by finding patterns you gain insight which allows you to find out not only what is happening but why, and then gives you the ability to make predictions and take action. “When I build AI solutions they usually have at least these three components – data, insight and action,” says Daniel.
Computers are usually very effective at finding patterns but not so good at adaptive decision-making and solving optimisation problems. This is because they understand what is happening but not why. “Computers can study masses of data and solve some problems in seconds, such as planning a delivery route. The point is that understanding is not what the computer does – people are needed to understand what the data means.
“And when we build the brain that is smarter than us it may remove us from the equation if we are not working together.”
If you give a machine the task of eradicating cancer it may well decide that to do that you need to kill all humans. We have to give machines the right objectives and constraints