The benefits for medicine are clear: “a more personalised prognosis and treatment plan.”
Koen Vermeer, Director at Rotterdam Ophthalmic Institute.
“I’m sure AI will begin to come into practice either this or next year.”
“Mostly useful for screening at first, then algorithms will replace existing algorithms – become more prognosis based.”
“Download models from the internet and, in a few hours, you have a working system.”
The risk?
"It’s going to be used before it’s fully understood.”
“Deep learning stuff is working really nicely – when there’s lots of data.” “With small improvements it’ll be good enough for actual use.”
What’s next?
“We’ll progressively see deep learning apply to more narrow use cases.”
“But I think we’ll see real progress when we integrate human or domain knowledge into these systems.”
“In lieu of data, we’ll use human guidance.”
“There’s no need for AI to learn all by itself.”
“For example, no need to learn gravity when we could just tell it.”
“The number of applications is much greater with augmented knowledge.”
Bio mechanical algorithm (brain) meets digital algorithm (computer).
“I think you could integrate those two things.”
“It’s like Google maps. Nowhere are all cars just simulated.”
“The real world is an analogue extension of digital computation.”
An interesting comment from the group:
"AI in healthcare system is a big step, which with the current development could take years. If we design a system that is artificially intelligent then there has to be a scope of self learning and improving over time. If we take out the self learning part, it no more an "intelligent" system. The usage of this term AI is no more appropriate in this case anymore, its just a controlled development of complex algorithm for which we humans are still responsible for the outcome. Are we at a stage that we can trust AI? I guess as engineers we are still busy addressing the aspect of what if AI goes rogue!"
Comments