You are currently viewing SemiWiki as a guest which gives you limited access to the site. To view blog comments and experience other SemiWiki features you must be a registered member. Registration is fast, simple, and absolutely free so please, join our community today!
I've been learning about AI and machine learning recently as they seems to be big buzz words. To my surprise, this isn't some revolutionary breakthrough, but that the theory and practice has been around for decades. It's really just recently become feasible with processing power improvements in the past few years.
From an investor perspective, is this winner take all?
For example, the popular AlphaGo program that plays Go from Google DeepMind. If someone where to try to build a program that could be beat AlphaGo would that be possible or is the head start to much to catch up?
There are really two separate aspects: the algorithm and the learning process. With AlphaGo they have been teaching it and simulating millions of games over multiple years. Now... is it possible to for someone to write a better algorithm that could overcome that disadvantage. How important is the learning process relative to the original algorithm?
Thank you. Just thought it might be an interesting topic to discuss.
We now track AI on SemiWiki and from what I have noticed it touches all of the other segments we track: Automotive, Mobile, IoT, and Security. It really is one of the broadest topics I have seen so I do not think there will be one winner, not even close. In fact, I think AI will touch just about every piece of silicon in the future. Security is one example. I see no other way in defeating the growing ranks of hackers now that hacking is a multi billion dollar industry. The good news is that AI consumes an amazing amount of silicon and not just memory but also SoCs. Everyone talks about the next killer app that will grow the semiconductor industry? That would be artificial intelligence.
Its worth noting that the CEO and Founder of Google's Deepmind has called Elon Musk an alarmist and sides with Zuckerberg. He's probably the most qualified of the three.
Its worth noting that the CEO and Founder of Google's Deepmind has called Elon Musk an alarmist and sides with Zuckerberg. He's probably the most qualified of the three.
I'm in between Elon and Mark. I do think the doomsday scenario of Elon with AI having mankind made to slaves or extinct is overblown. From the other side I do think - as with most new technologies - the regulations will need to evolve to make it fit for it.
Elon is right i'm afraid and so is Hollywood though they are underplaying the risk.
I'm reading this book Superintelligence: Paths, Dangers, Strategies, and true strong AI is minefield of bad outcomes.
Even if you don't agree with half of what this guy is saying much thought needs to go into supervising the path we are now on.
Elon is right i'm afraid and so is Hollywood though they are underplaying the risk.
I'm reading this book Superintelligence: Paths, Dangers, Strategies, and true strong AI is minefield of bad outcomes.
Even if you don't agree with half of what this guy is saying much thought needs to go into supervising the path we are now on.
I've been learning about AI and machine learning recently as they seems to be big buzz words. To my surprise, this isn't some revolutionary breakthrough, but that the theory and practice has been around for decades. It's really just recently become feasible with processing power improvements in the past few years.
From an investor perspective, is this winner take all?
For example, the popular AlphaGo program that plays Go from Google DeepMind. If someone where to try to build a program that could be beat AlphaGo would that be possible or is the head start to much to catch up?
There are really two separate aspects: the algorithm and the learning process. With AlphaGo they have been teaching it and simulating millions of games over multiple years. Now... is it possible to for someone to write a better algorithm that could overcome that disadvantage. How important is the learning process relative to the original algorithm?
Thank you. Just thought it might be an interesting topic to discuss.
Don't think it is fair to say there has been no big breakthrough. Neural nets in all their variants are relatively recent and a major departure from traditional algorithmic approaches to AI. Recognition rates in vision, voice and other areas were progressing only slowly until neural net methods were applied, when recognition rates skyrocketed, in a number of cases past human performance.
Should add that this area is still very new and fertile for new players. Far from a winner takes all market though good training databases in some domains are likely to be concentrated among some major players (FAANG)
General AI, AI that can think as well as a human :
If you believe the singularity scenario, even partially , about self-improving AI, AI could be a winner takes all.
But i think it is too early to tell, because we don't know how such thing would look like.
Deep Learning:
I think the algorithms are portable. So if both hardware platforms will be similarly capable(say like xilinx and altera are), it's no necessary a winner takes all.
But maybe Google, by attacking this in a fully integrated fashion - transistors to cloud services and developers could have some advantage and by continually investing it in improvement - would become a winner ? who knows.
Depends what you call relatively recent. I graduated mid nineties with a minor in AI; neural nets were already part of the curriculum then. I would rather call what is happening now a renaissance of AI.
It's true a lot of progress seems to be made in how to program these AI techniques and slowing of Moore's law scaling seem to force the microelectronic industry to diversify beyond pure compute improvement.
