Microsoft tries to write the book on AI
When it comes to the ethics and impact of artificial intelligence, Microsoft is literally trying to write the book. The Future Computed: Artificial Intelligence and its Role in Society is being made available in digital form tonight, with a forward from longtime lawyer Brad Smith and AI/Research chief Harry Shum.
Why it matters: Now is the time when a lot of key decisions are being made about how AI will work and the rules that will govern its development. But the discussion has largely been taking place within the tech sector.
Axios chatted with Smith and Shum about the book, AI in general and the questions we all should be grappling with. Below are the highlights:
So, why write a book?
Smith: There has been a lot of discussion in the tech sector, as you know. But we think that there's an important role to play in trying to take what the tech sector is talking about and broadening the dialogue.
Algorithms are already out there today helping make decisions on things like loans and jobs and parole. How fair and good is what's out there today?
Shum: I think we all understand that there's a lot of bias there in the existing AI algorithms. It's not even by any bad intention, it's just that there's a bias in the data, there's a bias in the model, there's a bias in the training. ... I don't think we today actually have the solution yet.
Ideally, we would like to have shipping criteria, that before we ship any AI algorithm systems, we actually pass those tests, and just like what we do today with accessibility, and we wouldn't ship those things until we passed the tests. But we're not there yet.
How big a deal is it that today it's overwhelmingly men, predominantly white men that are designing these systems? How big an issue is it at Microsoft, and how big for the industry as a whole?
Smith: I think it is an important issue for everyone. We all have to ensure that AI systems operate fairly. That means that it will be important to address the risks of bias creeping in to AI systems. I think it's fair to say that a more diverse industry and a more diverse company are likely to be more successful in designing AI systems that are free of bias.
Now, there are other ways to address bias as well, thankfully, but the faster we're able to move in creating an industry that is more diverse and more inclusive, the better off we're going to be in creating AI systems that are free of explicit or implicit bias.
One of the things you talk about in the book is the role of human judgment, not just in designing the systems but interacting with them and evaluating them over time. Is that something that's widely understood today or is this something that we need more of as an industry?
Shum: AI is here to amplify human ingenuity. We don't think that we develop AI to replace human beings. We believe that we are the first generation of humans ever living with AI. We need to learn how man and machine coexist and help each other.
I really believe that AI will help the humans to really do a lot of things better, and at the same time some skills AI will be able to do, already do better than us, and in the future, near future will be able to do better than us.
The third section in the book deals with how all of this impacts jobs. And you both talked in the book about the lack of a crystal ball, but what do you see as the most likely scenario playing out and over what timeframe? Who will the AI revolution be hardest on?
Smith: Let me just say the future won't be a good time to drop out of high school. If we think about the jobs that are most likely to be automated or eliminated because of artificial intelligence, they're most likely to be the jobs where computers can replicate the simpler aspects of human reasoning and the like.
Now, that's not necessarily new. The last three decades have not been a good time to drop out of high school either, but that's going to get worse rather than better.
But then the one other thing that I would say is there are a lot of causes for optimism. There will be many, many jobs that don't require a four-year college degree.
It does seem fair to predict that there will be a healthier dose of digital skills or data oriented skills or jobs than have been there in the past. So for all of us we're probably going to need to learn new areas of subject matter that involve digital skills.