Designing AI powered IoT: Thinking outside the app

We over promise in the short term, and under imagine in the long term

The first segment of Cockerill’s talk covered the short term over promising in contrast with long term under imagining. 

Cockerill said: “AI can enhance us rather than replace us. I think it’s interesting to think about artificial general intelligence, so that’s the idea that we can create a system that will replace or compete with human intellect and be able to do our tasks and that’s sort of massively ambitious.”

The kind of AI that is being talked about in these types of sweeping statements regarding AI replacing people is far off in the future, if ever truly reachable. These promises often seem far more informed by popular films involving AI than by the actual technology behind it. During this segment of his talk, Cockerill raised the point that he has no desire to be replaced, a sentiment which it is presumable many people share, so AI is likely to be developed to help and enhance, rather than replace.

Cockerill used the example of Google’s AlphaGo AI beating the world champion at Go, Lee Sedol, to illustrate how what may seem on the surface as AI becoming better than us at skills, but actually helps us to enhance our own abilities. Cockerill explained how after the defeat from AlphaGo, Sedol studied the techniques and strategies used by the AI to further enhance his own game, and proceeded to embark upon a long winning streak. 

From a design standpoint, Cockerill’s perspective on AI is therefore to think about where AI can be used to add real value to human experiences.

“There’s a paradox around artificial intelligence at the moment that the hard stuff is easy for AI, but some of the stuff we do that we think is easy, is hard for AI,” explained Cockerill, suggesting that because of this, AI will never be able to truly replace us, so our focus should instead be on developing it in such a way that the human and AI is blended together in the best possible manner. 

Cockerill goes on to compare AI as it is today as being in its childhood, and that it is our job to teach and mould it into the adult AI that is being promised, but is certainly not here yet. We don’t entrust AIs such as Siri or Alexa with critical tasks yet, as much like with a child, there is a high degree of chance that it will fail to complete the task, need help, or do something wrong. 

How can I connect with IoT, and what value can we bring if we design physical products?

Cockerill touched on deep learning with AI and systems learning from interactions, rather than algorithms and data fed directly into it, and knowing instinctually what is wanted from it, without needing to be directly asked. 

Looking outside of apps, which was also talked about in an earlier presentation from Ian Lee, Creative Director at Frog Design, and into hardware, is very important for the future of AI. Creating products that can use sensors and deep learning to begin to anticipate the wants and needs of people and carry out relevant tasks without having to be asked is far more useful and has far more scope than yet another AI app on your smartphone. 

Cockerill also highlights that every AI experience doesn’t have to be a ground breaking experience, it can simply be something that improves small things about your environment, such as the lights instinctually knowing when to turn themselves on. 

What experiences could AI enable?

As a designer, Cockerill is mostly concerned with where AI has the potential to go. He highlighted that way back in the sixties people were having the ideas that we see in smartphones today, but that they had no ability to create at the time. He emphasised that in that same vein it is important that designers today look to the future and don’t let the current restraints of technology trap them into under imagining the potential of AI.

Cockerill used three examples of AI that was enabled in a new way, or delivering different kinds of experiences to what AI commonly does today. The first was an AI that had been trained to recognise different objects, such as ties, which then might be able to anticipate someone’s intent based on recognising something it has seen, and provide what is wanted. 

Secondly he showcased an AI lighting sensor that could recognise the presence of a knife and bread, extrapolate from that the intention to slice bread, and thus project light onto the bread to leave a guide for the knife which adjust depending on the thickness of slices desired. Potentially this could be developed as a sensory lighting system to be able to measure out quantities and calculate cooking times as well. 

Finally Cockerill showed an AI system that interacts with other information, such as knowing a loved one is on a flight home, and then pulling in the readily available data pertaining to the flight. From there the AI was able to project that information onto a globe to track the progress of the flight and provide information the user might want, such as the hours until landing.

These are all examples of AI unobtrusively enabling and enhancing the activities, instead of trying to replace the person doing it. Cockerill’s talk concluded with the idea that systems with deep learning can provide richer and more individual AI, which can learn from its user to provide a fully personalised and optimised experience. So with that in mind, designers should turn their minds away from just catching data and placing it into a device, and instead be developing how to bring forward deeper IoT experiences.