Learning Library

← Back to Library

Balancing AI Power with Human Judgment

Key Points

  • AI’s potential goes far beyond large language models, yet many people mistakenly treat LLMs as the whole of artificial intelligence.
  • While LLMs excel at summarizing and synthesizing information, over‑reliance on them risks “dumbing down” human sense‑making and eroding our ability to develop judgment.
  • Our research program focuses on finding a balance between AI assistance and human reasoning, including creating audit tools to safeguard against loss of human capability.
  • Using LLMs for primary data collection can stifle imagination, abductive thinking, and the discovery of rare or out‑of‑pattern insights that only independent searching can reveal.
  • To preserve essential skills we must keep manual search, synthesis, and representation practices alive—much like climbing with a GPS, we need both technology and personal effort.

Full Transcript

# Balancing AI Power with Human Judgment **Source:** [https://www.youtube.com/watch?v=dvDGkP3V9Z8](https://www.youtube.com/watch?v=dvDGkP3V9Z8) **Duration:** 00:03:24 ## Summary - AI’s potential goes far beyond large language models, yet many people mistakenly treat LLMs as the whole of artificial intelligence. - While LLMs excel at summarizing and synthesizing information, over‑reliance on them risks “dumbing down” human sense‑making and eroding our ability to develop judgment. - Our research program focuses on finding a balance between AI assistance and human reasoning, including creating audit tools to safeguard against loss of human capability. - Using LLMs for primary data collection can stifle imagination, abductive thinking, and the discovery of rare or out‑of‑pattern insights that only independent searching can reveal. - To preserve essential skills we must keep manual search, synthesis, and representation practices alive—much like climbing with a GPS, we need both technology and personal effort. ## Sections - [00:00:00](https://www.youtube.com/watch?v=dvDGkP3V9Z8&t=0s) **Balancing AI Potential and Human Judgment** - The speaker emphasizes AI's broader capabilities beyond LLMs, illustrates how LLMs excel at summarizing content, warns that over‑reliance could diminish human sense‑making and judgment, and outlines a research program and audit tools aimed at preserving a balanced partnership between AI assistance and human decision‑making. - [00:03:08](https://www.youtube.com/watch?v=dvDGkP3V9Z8&t=188s) **Preserving Human Skills Amid AI** - A speaker warns that overreliance on language models can erode essential human abilities, urging the maintenance of those skills to navigate uncertainty. ## Full Transcript
0:01Okay. So the the first thing to say up 0:03front is that there is a lot of 0:05potential in AI AI and probably more 0:07potential in AI than just in LLMs and 0:10people tend to confuse the two. You know 0:13large language models are not the 0:14totality of what you can do with AI. 0:17Um and the utility is high. I mean this 0:20morning I was taking a transcript of a 0:22presentation I gave. I was taking some 0:24slides. I put that into an LLM and it 0:27produced a really good summary based on 0:28my original content which I can send to 0:30the client. So that ability to to 0:33summarize to synthesize is something 0:34LLMs are good at because they use 0:37improbability models based on training 0:39data sets. The real problem for human 0:42beings is that our own sense making 0:44capability could effectively be replaced 0:47by that. not by it doing what we do 0:50well, but it by dumbing ourselves down 0:52to the point where we actually don't see 0:54the capability of humans. So the big 0:57thing we're working on with the new 0:58research program is to get the balance 1:00between AI and and human based reasoning 1:03decision- making and to understand not 1:06only how that happens but the things 1:08that human beings have to do over an 1:10extended period in order to develop 1:12judgment type capability. 1:14The worry is the pace of change on LM's 1:17their use for human tasks could actually 1:19mean that we lose that capability or it 1:21gets lost substantially certainly from 1:23major companies and that's an issue 1:25hence our development of audit tools. 1:28Yeah. Um 1:30AI LLMs in particular taken unfettered 1:34and take for primary data collection are 1:37always going to reduce our imagination. 1:39They're going to reduce the 1:40possibilities. They think inductively we 1:42think abductively. We use metaphor. We 1:44use art. We use imagination. We actually 1:47use chemicals. We use senses and data we 1:50don't yet fully understand to make 1:52decisions under conditions of extreme 1:54uncertainty. 1:56Yeah. Where the past is not repeating in 1:58in a way which can be predicted by a 2:00framework or a model. So there are 2:02things and rules that we need to create 2:04around this. I blogged on this recently. 2:06Things like stop talking about AI as if 2:08it was a human being. Stop 2:09anthropomorphizing it. Yeah. actually 2:12make sure we practice things 2:13independently of LLMs. 2:16And certainly my own rule is not to use 2:18it for primary data collection or 2:20primary search because then I lose the 2:22capacity to find things I didn't expect 2:24to find. Remember looking through card 2:27indexes in the university library and 2:29finding a rather obscure essay by a guy 2:31called Joseph Stalin when he was a 2:33doctoral student at a college in Georgia 2:35on toolsto what is art. Now that would 2:37have never come up with an LLM because 2:39it was an outlier. it wasn't a dominant 2:41pattern. So we need to maintain search 2:44capabilities. We need to maintain 2:45production capabilities, synthesis and 2:48representation. I can see AI doing a lot 2:50more of that. But again, we need to 2:52maintain our skill base. And probably 2:55the best metaphor I can think of is I 2:57make sure I climb a lot. I walk a lot. 2:59I've always got GPS. I've got two GPS 3:01systems. But about once a month, I make 3:04sure I walk with a map and compass. 3:05Otherwise, I will lose that skill. So 3:08start to think about the skills that 3:09human beings have they need to maintain 3:12which while LLM might help with that, 3:14augment it or act as a hybrid, we have 3:16to maintain those skills for conditions 3:18of uncertainty. 3:22[music]