Name: Insights: IBM On How AI Is Already At Work
Uploaded: 2020-02-25
Duration: 3 min 58 s
Description: Dario Gil, IBM's VP of A.I. and quantum computing, breaks down what we saw in 2018, and what we'll see in 2019 and beyond. (December 2018)

[MUSIC PLAYING] DARIO GIL: I think what's happened right now is that AI has really begun to come into our lives. And we've begun to see devices like a smart speaker where now we have a speech interface. And underneath there will be AI models that are implementing the speech recognition and then we can do simple functions. Or we may have in our cars some form of intelligent assist. To give one last example, better weather model prediction. Is it going to rain? Or will they have also applications in yield prediction in agriculture? We've seen the increased use of machine learning models behind the scenes. So in many different aspects of our lives, a narrow form of AI has begun to work and be present. One of the most exciting ones in the field of AI is something called Project Debater. So we have been working for the last six years within IBM Research on creating an AI system that can engage in a full live debate with a human being from one sentence-- not one sentence, one comment, one sentence, 10 minutes of content that also involves some amount of the listening, comprehension, and the rebuttal. That to me is an amazing achievement that is indicative of what is the next phase for AI, going from a narrow AI to a broader form of AI that is more capable and more sophisticated in the use and understanding of language. If we paint the broadest possible picture, we can envision a future in which we humans, we're going to have a complementary expertise, as we go on tackling making decisions. Today, we form institutions, companies of different forms, where through the expertise of the people who work there, we work with each other collaboratively to solve problems. Why couldn't we have a form of AI that is part of those colleagues? Definitely, we're making advances in making AI easier to use. One key area is that as we make more progress on technologies like transfer learning and bringing in reasoning along with learning, the amount of data that is required to train the system goes down. A second one is around the automation of AI. It is becoming easier and easier to have machines to automatically construct both the neural network and increasing amounts of the data science associated with training these systems as well. The computation for AI is also improving quite dramatically. So you can train more efficiently at lower cost and faster. We are going to see increasing democratization of the ability to create AI, and very importantly, the ability to deploy AI, and make sure that it is trusted, which to me is the big, big new 2019 center stage is going to be the issue of not just accurate AI, but trusted AI. As we think about AI, learning is not enough. We also need reasoning in addition to learning. Because it allows us to create also models of the world and think about possibilities we haven't yet encountered or for which data is not there. When you actually look into the details of what that would mean, it is likely to be the basis of a significant leap in reading comprehension. And if we could actually build AI systems that comprehended text, and books, and so on, that would be a significant and very important milestone for the field that would utter the advances of creating these systems that can really complement our knowledge and our expertise. Just like in the Industrial Revolution, we extended what our muscles could do through machinery. Now, we would truly have a companion to help us expand what our knowledge, and our reasoning, and our decision making can be, but it starts with unifying learning and reasoning. [MUSIC PLAYING]

Insights: IBM On How AI Is Already At Work

Uncategorized

IBM's Dario Gil on what we'll see in 2019 and beyond.

IBM On How AI Is Already At Work

Dario Gil, IBM's VP of A.I. and quantum computing, breaks down what we saw in 2018, and what we'll see in 2019 and beyond. (December 2018)

I think what it happened right now is that a I has really begun to come into our lives on. We begun to see devices like Smart Speaker, where now we have a speech interface and underneath there will be a I models that are implement in the speech recognition and then weaken the simple functions. Or we may have in our cars some form of intelligent assists to be one last example. Better weather models prediction. It's gonna rain or it will be handled. The implication in like your prediction in every culture, we seem to increase use machine learning model behind many different aspects of our lives. A narrow form of a I has begun to work the press one of the most exciting ones in the field of a I from single project debater. So we have been working for the last six years within IBM research on creating a system that can engage in a food life debate with a human being for one cent, not 1/10 of one command, one sentence, 10 minutes of content that also involved some amount of like the listening comprehension under a bottle. That is an amazing achievement that is indicative. What is the next Phase II going from a narrow way, I tow a broader formally I that is more capable and more sophisticated use of language, way paint, their broadest possible picture. We can envision a future in which we humans we're gonna have a complimentary expertise as we go on attacking, making decisions. Today we form institutions, companies of different forms where who with expertise off the people who work there. We work with each other collaboratively to solve problems. Why couldn't we have a form of the eye that is part of those colleagues? Definitely. We're making advances in making a I easier to use. One key area is that as we make more progress on technologies like transfer learning on bringing in reasoning, along with learning the amount of data that is required to train this system goes down. The 2nd 1 is around. The automation of it is becoming easier and easier to have machines to automatically construct both the neural network and increasing amounts of the data science associated with training. These systems as well the computation for a I. So it's improving quite dramatically so you can train more efficiently at lower cost and faster. We're gonna see increasing democratization off the ability to create a I and very importantly, the ability to deploy a make sure that it is trusted, which to me is a big, big new 2019 center stage is gonna be the issue of not just accurately I trusted me. I way think about a I learning is not enough. We also need reasoning in addition to learning, because it allows us to create also models of the world and think about possibilities we haven't yet encounter or for which data is not there. When you actually look into the details of what that would mean, it is likely to be the basis of a significant lead in reading comprehension. I mean, we could actually really I system that comprehended tax on books and so on. That would be a significant, very important milestone for the field that would utter their advances. Off created systems that can really compliment or knowledge or expertise, just like in the Industrial Revolution. We extend that what are muscles could do machinery. Now we would truly have a companion to help us expand what our knowledge reasoning on our decision making candy. But it starts with define learning and reasoning