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!
With new forms of AI/ML coming online and the amount and types of AI/ML increasing, what impacts can we expect to see on the industry and society? Will data centers proliferate with each having their own specialty and how will society handle changes to the labor market that will go from decades to years? Will these tools feed an ever accelerating rate of change with compounding resulting in a knowledge and application explosion that will disrupt everything as it feeds on itself. It is easy to see massive displacements in the labor force at all levels as automation geometrically displaces both the physical and mental pursuits in a decade that have been with us for centuries. Will this lead to a concentration of wealth and power or a demonization of wealth and power? Will application specific AI/ML or general AI/ML chips be the future or both. Will data centers or distributed AI/ML or both be prominent? The semi sector is now set to make changes society has never dealt with before as our creations surpass our own. Any thoughts or comments appreciated.
With new forms of AI/ML coming online and the amount and types of AI/ML increasing, what impacts can we expect to see on the industry and society? Will data centers proliferate with each having their own specialty and how will society handle changes to the labor market that will go from decades to years? Will these tools feed an ever accelerating rate of change with compounding resulting in a knowledge and application explosion that will disrupt everything as it feeds on itself. It is easy to see massive displacements in the labor force at all levels as automation geometrically displaces both the physical and mental pursuits in a decade that have been with us for centuries. Will this lead to a concentration of wealth and power or a demonization of wealth and power? Will application specific AI/ML or general AI/ML chips be the future or both. Will data centers or distributed AI/ML or both be prominent? The semi sector is now set to make changes society has never dealt with before as our creations surpass our own. Any thoughts or comments appreciated.
In my opinion: AI/ML mostly being software (enabled by underlying compute silicon), and everything else looking down economically, means much of any remaining glut of VC/F100 dollars will be trying to crowd a smallish space (writing in terms of proven solutions) that have even smaller actual applicability when it comes to high-quality outcomes. Bubbles will burst, some fortunes will be made.
Forgive me for saying so, but whether it's ChatGPT or MidJourney, they aren't 'some kind of dark magic'. It's just advanced statistics performed across huge datasets. The quality of such outputs is "cool at first" but highly superficial. Actual meaning or information in this class of outputs remains low. And I'm concerned across use cases that it could limit creativity by not being capable of going beyond what the old corpus of data may train it for.
Past applications in the same vein, e.g., interactive voice response, or automated business processing, or filing US taxes (I won't go on), are STILL painstakingly difficult in spite of billions and decades spent chasing a dream of not needing humans. Perhaps the current wave of new and improved offerings will be worked into better system queues to pre-process something a human needs to look at and action...but left on their own, these systems would run amuck (maybe trying to make your tax return look like some sort of random mid-sigma median of US averages).
For semiconductors, AI/ML can probably have a meaningful impact in making EDA and other design tools function faster as part of a wholistic process, combined with more precise software tools for validating produced designs. This could have a big impact in IP specifically. If AI/ML can quickly thrash about to produce a design with similar input/output mappings as a competitor's product, it's not IP theft, right? What I don't know, is would that be efficient? Would you be spending more cycles trying to copy (the CCP way?) than just going for a solid design?
My few days of experience with the new ChatGTP found the biggest difference between "it" and me is that "it" lacks curiosity. "It" is always ready to answer (almost) any questions that I throw at "it", with its vast library of information, but it does not have any real "intelligence" because it just keeps answering and answering, but have no capacity to "think". It never challenges me why I ask the same question twice or different ways three times, what am I think, or what may be the end game I'm trying to get to, etc. As of today, it's just a (much) better dictionary, one cannot have a conversation with "it" yet. Long way to go to be worthy the label of "intelligence".
In many areas, so called professionals do what current AI/ML does, but very poorly and on a smaller scale. The ability to construct an ever more powerful data base as more input is added in a far more economic manner than teams and people is just plain math and economics. Forty percent of a doctors time is spent on diagnostics and it would be better for a large data base to deliver more accurate results and give the doctor a far better range of choices. Modern tools can see color, heat and texture plus real time variations on the body far better than any human can in a small fraction of the time and corelate these factors over large historical data bases. Many professions are about protecting themselves than serving the public.
