Artificial intelligence is worse than humans in every way at summarising documents and might actually create additional work for people, a government trial of the technology has found.
Amazon conducted the test earlier this year for Australia’s corporate regulator the Securities and Investments Commission (ASIC) using submissions made to an inquiry. The outcome of the trial was revealed in an answer to a questions on notice at the Senate select committee on adopting artificial intelligence.
The test involved testing generative AI models before selecting one to ingest five submissions from a parliamentary inquiry into audit and consultancy firms. The most promising model, Meta’s open source model Llama2-70B, was prompted to summarise the submissions with a focus on ASIC mentions, recommendations, references to more regulation, and to include the page references and context.
Ten ASIC staff, of varying levels of seniority, were also given the same task with similar prompts. Then, a group of reviewers blindly assessed the summaries produced by both humans and AI for coherency, length, ASIC references, regulation references and for identifying recommendations. They were unaware that this exercise involved AI at all.
These reviewers overwhelmingly found that the human summaries beat out their AI competitors on every criteria and on every submission, scoring an 81% on an internal rubric compared with the machine’s 47%.
As someone who regularly uses AI for my podcast, it’s absolutely worse in every way, but it does 95% of the work. Creates chapters, makes subtitles and a summary, and then I come and just clean up the rest. If it weren’t for AI I just wouldn’t bother with it.
I think this is the thing a lot of people don’t understand about AI; it’s not going to replace humans entirely, but it can make a human way more efficient, and make 1 human able to do the work of 3-5 humans.
I’m doing a series of conversations/interviews with my parents’ generation to keep a voice record of their stories. As part of that, I’m doing transcripts that start with the transcript feature of Google’s Recorder. It can do some nifty things like assign speakers to individual voices. I have to clean up the transcripts some, but it’s far less laborious than dealing with a 15-20 minute conversation. I can fix up a transcript in maybe 5 minutes.
Exactly
Not if you have to proof-read everything to spot the entirely convincing-looking but completely inaccurate parts, is the problem the article cites.
You think it’s not faster to read and correct than it is to write the whole thing?
If the error is hidden well, yes. Close-reading a text and cross referencing everything it says takes MUCH longer than writing a piece you know is accurate to begin with
Absolutely incorrect.
For summarization, having the data correct is crucial because manual typing itself is not a large chore. AI tends to shine more when you’re producing a lot of manual labor such as a 10-page document for something. At that point, the balance tips the other way where proofing and correcting is much easier and less time-consuming than the production itself. That’s where AI comes in for the gains in workflows. It has other fantastic uses as well, like being another voice for brainstorming ideas. If done well, you’re not taking the AI’s idea so much as just using it to spur more creative thinking on your end.