The basic idea is very simple. Right now, when you want to know about something, you start by doing a search. You get a bunch of snippets and links to likely documents. So you go to the best-looking ones and read. If your question is fairly broad you might read quite a bit by way of piecing together an answer. Your computer does the search then you do the research.
Suppose AI does the research and reports back to you with your answer? This seems perfectly possible; in fact, it is what these new AI engines seem to do best.
When you think about it, we do a lot of research.
Shopping likely tops the list. So a query might be to find me the cheapest something within a specified distance. A specific car within 300 miles, say, or a special pizza within 5 miles. We do that a lot. Or the best price in all the online outlets, which would take me endless time.
Then there is looking over time. Most stores have sales off and on. For example, Amazon prices for a given item can change a lot over time, including being briefly what looks like erroneously low.
Your AI shopper could effortlessly spend what would be all your time spotting sales. You could even have low ball buy orders just like automated stock buying software does today, but with much more complex instructions from you.
Then, too there is reading product or service reviews and comments. Online reviews are a wonderful addition to shopping, but they are time-consuming to read and digest. AI could do this, but maybe not today’s systems. The problem is that people disagree. Some like it, some do not, and everybody has reasons.
Which brings us to a key issue, namely how does AI handle disagreement? Much of our daily research involves assessing conflicting opinions, but I have yet to see a discourse AI product that can do this well, or even at all.
Much of what we want to know about is controversial. Products, services, politics and policy, sports, diets and health, child rearing and education, news, science, law and regulation, etc. Today’s AI systems seem to take positions based on their training rather than analyzing the issues for us. This is a serious weakness.
Maybe issue analysis will be the next big thing in AI. It may take new programming or, training, or both because issues have a distinct structure. See my “The structure of complex issues” here: https://www.cfact.org/2020/11/17/the-structure-of-complex-issues/. It could also track issues as they evolve over time.
Getting back to the basic point, the WorldWideWeb just turned 30 years old. It has made search and research a key part of daily life. The computer does the search, and we do the research on those search results. There are two kinds of research — fact-finding and issue analysis.
If AI can do some of that research, it could be a major change in our lives. However, that change will be electricity intensive. Some projections of AI power usage are hard to believe as they range up to a whopping 20% of American electricity consumption.
I have not looked at these AI growth models, so have no idea how realistic they are. It is likely they are way too fast. A technology is more than stuff; it is people using stuff, which takes a long time to develop. If AI can do some of our research, it will definitely grow big.
If AI can do our basic research, that would be extremely useful. It would free us up to do other things, like thinking and more advanced research, or even relaxing a bit. Stay tuned.
David Wojick, Ph.D. is an inDr. David Wojick is an independent policy analyst and senior advisor to CFACT. As a civil engineer with a Ph.D. in logic and analytic philosophy of science. This article was first published HERE
When you think about it, we do a lot of research.
Shopping likely tops the list. So a query might be to find me the cheapest something within a specified distance. A specific car within 300 miles, say, or a special pizza within 5 miles. We do that a lot. Or the best price in all the online outlets, which would take me endless time.
Then there is looking over time. Most stores have sales off and on. For example, Amazon prices for a given item can change a lot over time, including being briefly what looks like erroneously low.
Your AI shopper could effortlessly spend what would be all your time spotting sales. You could even have low ball buy orders just like automated stock buying software does today, but with much more complex instructions from you.
Then, too there is reading product or service reviews and comments. Online reviews are a wonderful addition to shopping, but they are time-consuming to read and digest. AI could do this, but maybe not today’s systems. The problem is that people disagree. Some like it, some do not, and everybody has reasons.
Which brings us to a key issue, namely how does AI handle disagreement? Much of our daily research involves assessing conflicting opinions, but I have yet to see a discourse AI product that can do this well, or even at all.
Much of what we want to know about is controversial. Products, services, politics and policy, sports, diets and health, child rearing and education, news, science, law and regulation, etc. Today’s AI systems seem to take positions based on their training rather than analyzing the issues for us. This is a serious weakness.
Maybe issue analysis will be the next big thing in AI. It may take new programming or, training, or both because issues have a distinct structure. See my “The structure of complex issues” here: https://www.cfact.org/2020/11/17/the-structure-of-complex-issues/. It could also track issues as they evolve over time.
Getting back to the basic point, the WorldWideWeb just turned 30 years old. It has made search and research a key part of daily life. The computer does the search, and we do the research on those search results. There are two kinds of research — fact-finding and issue analysis.
If AI can do some of that research, it could be a major change in our lives. However, that change will be electricity intensive. Some projections of AI power usage are hard to believe as they range up to a whopping 20% of American electricity consumption.
I have not looked at these AI growth models, so have no idea how realistic they are. It is likely they are way too fast. A technology is more than stuff; it is people using stuff, which takes a long time to develop. If AI can do some of our research, it will definitely grow big.
If AI can do our basic research, that would be extremely useful. It would free us up to do other things, like thinking and more advanced research, or even relaxing a bit. Stay tuned.
David Wojick, Ph.D. is an inDr. David Wojick is an independent policy analyst and senior advisor to CFACT. As a civil engineer with a Ph.D. in logic and analytic philosophy of science. This article was first published HERE
No comments:
Post a Comment