The international audience at the recent World Summit AI 2017 in Amsterdam, The Netherlands, reportedly topped out at 2,400. Not bad for an inaugural event, writes Jeremy Cowan, editorial director of IoT Now. And it was the goal of a panel of AI heavyhitters to inspire. Judging by the attentive audience, they succeeded
Amazon’s Ralf Herbrich joked that it might be quicker to say where artificial intelligence (AI) is not impacting Amazon’s business. He cited current uses of AI in areas as diverse as demand pricing, fresh fruit ripeness prediction, and in Alexa.
Rob High chipped in saying that IBM Watson uses AI to help doctors identify the right time for oncology (cancer care) treatments, and “we allow other people to take AI into other areas such as conversational systems” or deeper insights into retail customer behaviour.
If those are some of the areas in which AI is working now, said the moderator, Prof. Max Welling, what could AI do better?
We need more accuracy, less energy – Herbrich believed we can use AI to play games with humans but that’s not presently a good use of computing power. A video game would involve a human burning energy at the rate of 2,000 calories per day, while it would take 220 million calories for AI computing to do the same job. “So we have to reduce energy,” he said. “We need more accuracy per calorie, which means compressing neural networks.”
If that’s so where are the bottlenecks in AI, the moderator wondered.
Tencent’s Zhang Tong pointed to dialogue systems, “as an industry we’re not doing very well in natural language or robotics, how we take actions with AI. There’s room for improvement.”
As to where AI and machine learning (ML) can best help society, High felt, “cognitive computing and AI should be there to amplify our thinking. The danger isn’t simply that it might rise up and take over our role. There are things we do well, but there are a lot of things we don’t do well. AI can extend our reach, focusing on what gives us the biggest opportunity to enhance ourselves. We need to see through our biases to allow us to make better decisions.”
Asked if humans and AI are co-evolving, High said, “Yes, it’s a complement to human intelligence, (AI) will co-reside with us.”
AI as a team sport – Marco Vernocchi, senior managing director, Applied Intelligence lead at Accenture commented, “AI is a team sport. No single organisational unit or discipline can make it a success. Also, it’s an ecosystem play.”
Herbrich liked the team sport description. He went further saying, “For many years the academic community lacked the ability to know what AI’s applications will be. It was an unguided field but in AI industry and academia meet well.”
Prof. Welling was concerned to know if the right skills were available for AI to develop. Aren’t people in academia (with AI skills) being acquired by industry?
Zhang Tong was quick to scotch these concerns. “I was a professor at Rutherford University before joining Tencent. I see it as a benefit not a problem. There is not a lack of talent, there are very strong people who will build their own teams. We collaborate with universities in the US, Europe and China. It’s a healthy environment and is sustainable.”
People often talk about artificial intelligence in terms of its threat. “What is the threat?” the panel was asked.
High replied, “Not paying attention to how these technologies are affecting us. Think about smartphones, they’ve changed our behaviour. Quite subconsciously AI will change us in profound ways but also subtle ways, and we’ll need to be aware. For example, voice recognition is so effective we almost don’t notice it any more. Do we recognise it as AI?
Vernocchi agreed. “Predictive text and autocorrect have affected us. We now type stuff that way because we know it will be autocorrected. My parents would never have typed that way.”
He concluded, “There’s a big gap between the very fast speed of technological development and the speed of regulation (change). So there’s a development responsibility on all of us.”
The event heard later from Gary Marcus, founder and CEO of Geometric Intelligence (acquired by Uber), that a lot of what we might want to do with AI remains out of reach.