One of the most interesting sessions at the recent Dreamforce conference was the fireside chat with IBM CEO Ginni Rometty, an hour long discussion that covered a lot of ground. In the mix were some compelling views on AI, both in terms of its potential for good, but also of the responsibilities that came with such new tech.
Time constraints restricted the amount of discussion that could take place at Dreamforce, so it was good to see many of the topics and concerns alluded to by Rometty picked up in more detail by Doctor John Kelly, Head of Cognitive Solutions and IBM Research, at the 21st Annual Credit Suisse Technology, Media & Telecoms Conference this week.
Cognitive is IBM’s preferred terminology for a range of tech that covers AI, machine-learning et al. Kelly took time out to define some of IBM’s nomenclatures and categorisations, beginning with Narrow AI:
This is machine learning to simply identify patterns of things. In the case of imaging, lots of companies use very narrow AI to sort through the pictures that you store on the cloud or on your phone, as an example, to identify family members or cats or pets or whatever, or an industry to find defects in products like pharmaceuticals, that’s very narrow, reinforced learning technology.
Then there’s General AI, he said, which is:
The infamous Black Box that can do anything.[It can] build its own models, it can reason, it can answer questions on anything. That is decades away. Everyone in the industry believes that that is a long way away.
Today though there is a big space left between those two whch can be catgorised as Broad AI:
This is multi-modal data, it’s complex data, it’s complex processes, whether it’s oil rigs or healthcare, financial services, this is big complex data. But this is where the value will be and will be extracted in the next several years through the next couple of decades and this is where we have focused our business in this area.
Into the business
Kelly observed that the shift to cognitive computing has been accompanied by a shift away from IBM’s traditional comfort zone of the CIO and back-office data processing, out into the line of business:
We may be within the same company, but we're talking to entirely different people. We're not to talking to the CIO of a hospital, we’re talking to the Chief Medical Officer. He trusts us because they know we run all their backend systems. But we have to earn the right then to add value to that business line.
What we find is we always start with POCs, proof of concept. They can be small or large. They’re very quick now, because we can train Watson our new data extremely fast. These are now POCs that last week's, not months. We've demonstrated value and then it quickly turns to, ‘OK, how do we scale this very quickly in the enterprise?’ and then they are very quick to say, ‘OK, I have three more use cases, which ones are we going to do next?’.
Cognitive customers are coming from across the spectrum of industries, he added, with financial services as a case in point “across the board”. But it was to healthcare that Kelly returned several times in his dicussion, as Rometty had herself done at Dreamforce:
Healthcare is an interesting one, because we're finding that this is more of a developing market. So we developed the capability in US and we’ve got lots of clients in the U.S., hospital systems, using Watson for oncology, as an example.
But it’s not just a US phenomenon, he added:
I was with the CEO of a big hospital system in India, Manipal Hospital two weeks ago. He said ‘John, we have 1.3 billion people in India, 300 million of them roughly can afford to see an oncologist. God forbid they have cancer’. [There is] at least a billion people that have no access to an oncologist. He said then within that 300 million, he said, ‘We only have 700 or 800 oncologists in the whole country and we can't serve that 300 million people who can afford it, much less the billion’. He said [he] couldn’t build enough hospitals and medical schools to produce the oncologist to serve even that population.
And so he said, ‘I see Watson oncology as a way to up the game across my entire healthcare system and bring the best training in the world from the US, the use case of the world there in seconds and then up the productivity of the 750 oncologists I have and perhaps even take some of the GP practitioners and raise them to the point where they can start to offer enhanced healthcare around cancer’. So it was stunning for me and we're seeing the same thing in China where it is just growing exponentially in terms of the adoption in healthcare…where there is the big knowledge gap, we're finding the fastest adoption in big populations.”
That’s the benefits side of the cognitive revolution, but Rometty was particularly empassioned about the responsibilities as well, particular around ownership and use of the exponentially-increasing levels of data generated today. Kelly picked up this theme:
It has moved from two years ago [when] our clients weren’t even thinking about it. They knew they had a lot of data, but they weren’t thinking about what to do with it and what not to do with it. About a year ago, they started to really hear about the data and what do [they] do with it? At the same time, a lot of companies were coming to them saying, hey put your data on my cloud in fact, extremely low price or bring it for free. I'll give you free storage capacity. And they started to really think about the implications it had.
Over the last six months or so, I’ve seen a pretty dramatic shift to where enterprises realize the value of what they have in that data. Whether you’re a pharmaceutical company and the value of the data from your either successful or unsuccessful clinical trials is gold. Yesterday, I was with a big pharmaceutical company CEO and there is no way he is going to put that clinical trial data on a cloud where that company is building a large AI model of human behavior and healthcare outcomes, and have those models then worked with the competitors, as an example.
IBM gets this, he insisted:
In our case, we’ve been by way very open about our strategy around data. Number one is we tell every one of our clients, ‘Your data is your data and if you put on our systems, we will protect it with our lives’. That’s just the way we’re wired. Number two is if we use AI, we agree [with you] to use AI and we will tell you every time we use it. And every insight that comes out of that use of AI and your data is your insight, your unique insight. We will not let that insight on a drug or on a person get to another company. And then lastly we’ll be transparent in what forms of AI we use. We will show you the algorithms. We will be completely transparent. I don’t know if any other company has taken those three positions.
All of this is particularly important in light of the type of customers that IBM deals with, said Kelly:
We are the enterprise company of the world…and 80% of the world’s data is held by those companies. Only 20% of the world’s data is accessible on the Internet to the search companies of the world. So we have access to 80% of the world’s data and it’s very domain specific to those industries.
Looking further afield, the next step is towards Quantum Computing, something which Kelly admitted he had not expected to come to pass in his lifetime:
Five years ago, I said I'm going to see one in my life. I now see one in my career. We have put a 50 qubit system up. I have demonstrated 50 qubit system. We have 17 qubit systems on the Web you have to play with, only company that’s been able to do that. So here comes another exponential law in computing that is going to be huge and disruptive.
The emergence of workable quantum solutions will have enormous practical benefit, explained Kelly:
You can do incredible things from a security standpoint, encrypting and decrypting. In pharmaceuticals, you can model from the ground up energy states, new molecules for drug discovery. You can go into supply chains and do optimization routines at speeds you can't touch for a classic computer. In financial services, you can do optimization of things like trading at speeds that even the largest supercomputers in the world can't do. So the race is on to build that two scale quantum computer. There is only a few of us in the world that are doing it or capable of doing it today. We’re in the race.
An interesting and engaging discussion from a man with a clear passion for the potential of technology. As he said:
I was early-on on the Moore's law curve and built much of my career in Moore's law and Metcalfe’s law and I'm fortunate enough to be around for the early part of the AI’s law.