Last week's conversation with Allison Pugh painted a rather terrifying portrait of the age of Artificial Intelligence and its continued gutting of the human workforce, from grocery store clerks to medical assistants to novelists.
Gosh, I find myself in the weird position of arguing against good people who stand for the right things. How can that be?
Well, the issue here is that a bunch of ideas are being conflated that should really be kept separate. And as a consequence, some babies are being partially thrown out with the proverbial bathwater. Let me explain.
Broadly, Emily and Alex are challenging a few things:
1) The BUSINESS of AI and AI startups
2) The TECHNOLOGY of AI and specifically LLM AIs
3) The ETHICS of how AI has gathered their databases
4) The idea that AI can REPLACE CREATIVE work
5) CAPITALISM and its drive to increase shareholder/owner value at the expense of workers
Let's take these on one at a time.
First, AI startups are indeed likely an inflated bubble that will, at least partially pop. We've watched this movie before. Remember the Internet bubble and its eventual popping?
HOWEVER, remember that the Internet itself wasn't invalidated as a technology nor did all Internet businesses die. Witness Amazon and Google as just two examples. Similarly, the idea that AI will die and go away is simply a non-starter. It is a useful tool, when used properly (more on that later).
Second, Emily and Alex (and you) are correct that LLM AIs are not "reasoning". Ironically, the pursuit of real, "reasoning" AI probably slowed down the development of USEFUL AI until the LLM idea was explored. But now we might just be arguing semantics. Do LLM AIs pass the Turing test? In many, many instances, they do. More importantly, does this approach yield useful results? I argue yes, when used within its limitations.
As an example, programmers can get a start on a coding task that saves them tons of time. BUT! these draft solutions are just that. Drafts that must be examined and tweaked by a knowledgeable human.
A rough analogy - lawn care. AIs can mow your lawn. You still have to come around with the trimmer to finish the job.
Third, the ethics (or lack thereof) of how LLMs scraped the Internet for its sources. This is absolutely 100% a crime. No argument here. Over time I expect to see AIs get certified for ethically gathered source material not unlike how coffee beans are certified Fair Trade.
Fourth, replacing creatives with AI. I agree with you all here. No one should be expecting AI to replace real artists, writers, designers, etc. You will get, at best, "meh" work. Or highly derivative work that is frustratingly mediocre.
My only slight quibble here - humans are nearly as prone to deliver work of this kind for a variety of reasons. Nonetheless, I think this brings us back to my point that AI should be used within its understood limitations.
I wouldn't write a paper with it. But I might ask it for a bulleted list of items relating to my topic to help ensure I haven't missed something. Did that with my daughter a couple of years ago vis a paper on climate change she was writing. And indeed, it listed something she'd meant to cover but forgotten.
Finally, Capitalism. Without falling into the basic "capitalism is bad" stance, I have to agree that the drivers of capitalism will propel us down paths that hurt the folks at the bottom of the pyramid (and those in the middle, increasingly). And watching naive managers jump on the AI bandwagon as a thinly veiled means to rid themselves of those pesky (and costly) workers is infuriating on many levels. Not least of which is that we will all wind up being hurt by this.
But once again, shorn of its worst impulses, the urge to find ways to use better tools to deliver better results is not wholly off base. Robots perform precise, repetitive tasks better than humans. AI data analysis can highlight issues that humans might overlook. Again, we have a powerful tool that used correctly with proper goals can be an amazing boon to all of us.
So my frustration here is that the idea being pushed is AI bad AND AI fraudulent. But I don't think that's particularly useful. PEOPLE are often bad and often do things with the wrong goals in mind. And they'll use whatever tools come to hand to further those bad intentions. The tools themselves aren't necessarily the issue.
Gosh, I find myself in the weird position of arguing against good people who stand for the right things. How can that be?
Well, the issue here is that a bunch of ideas are being conflated that should really be kept separate. And as a consequence, some babies are being partially thrown out with the proverbial bathwater. Let me explain.
Broadly, Emily and Alex are challenging a few things:
1) The BUSINESS of AI and AI startups
2) The TECHNOLOGY of AI and specifically LLM AIs
3) The ETHICS of how AI has gathered their databases
4) The idea that AI can REPLACE CREATIVE work
5) CAPITALISM and its drive to increase shareholder/owner value at the expense of workers
Let's take these on one at a time.
First, AI startups are indeed likely an inflated bubble that will, at least partially pop. We've watched this movie before. Remember the Internet bubble and its eventual popping?
HOWEVER, remember that the Internet itself wasn't invalidated as a technology nor did all Internet businesses die. Witness Amazon and Google as just two examples. Similarly, the idea that AI will die and go away is simply a non-starter. It is a useful tool, when used properly (more on that later).
Second, Emily and Alex (and you) are correct that LLM AIs are not "reasoning". Ironically, the pursuit of real, "reasoning" AI probably slowed down the development of USEFUL AI until the LLM idea was explored. But now we might just be arguing semantics. Do LLM AIs pass the Turing test? In many, many instances, they do. More importantly, does this approach yield useful results? I argue yes, when used within its limitations.
As an example, programmers can get a start on a coding task that saves them tons of time. BUT! these draft solutions are just that. Drafts that must be examined and tweaked by a knowledgeable human.
A rough analogy - lawn care. AIs can mow your lawn. You still have to come around with the trimmer to finish the job.
Third, the ethics (or lack thereof) of how LLMs scraped the Internet for its sources. This is absolutely 100% a crime. No argument here. Over time I expect to see AIs get certified for ethically gathered source material not unlike how coffee beans are certified Fair Trade.
Fourth, replacing creatives with AI. I agree with you all here. No one should be expecting AI to replace real artists, writers, designers, etc. You will get, at best, "meh" work. Or highly derivative work that is frustratingly mediocre.
My only slight quibble here - humans are nearly as prone to deliver work of this kind for a variety of reasons. Nonetheless, I think this brings us back to my point that AI should be used within its understood limitations.
I wouldn't write a paper with it. But I might ask it for a bulleted list of items relating to my topic to help ensure I haven't missed something. Did that with my daughter a couple of years ago vis a paper on climate change she was writing. And indeed, it listed something she'd meant to cover but forgotten.
Finally, Capitalism. Without falling into the basic "capitalism is bad" stance, I have to agree that the drivers of capitalism will propel us down paths that hurt the folks at the bottom of the pyramid (and those in the middle, increasingly). And watching naive managers jump on the AI bandwagon as a thinly veiled means to rid themselves of those pesky (and costly) workers is infuriating on many levels. Not least of which is that we will all wind up being hurt by this.
But once again, shorn of its worst impulses, the urge to find ways to use better tools to deliver better results is not wholly off base. Robots perform precise, repetitive tasks better than humans. AI data analysis can highlight issues that humans might overlook. Again, we have a powerful tool that used correctly with proper goals can be an amazing boon to all of us.
So my frustration here is that the idea being pushed is AI bad AND AI fraudulent. But I don't think that's particularly useful. PEOPLE are often bad and often do things with the wrong goals in mind. And they'll use whatever tools come to hand to further those bad intentions. The tools themselves aren't necessarily the issue.