I've been reading Ethan Mollick's new book "Co-intelligence: Living and Working with AI" It's an excellent read for anyone interested in the topic; there are some gems in it. One of them is how creative AI is. That's something I've seen, for example, when I drew a picture based on a poem by Wilfred Owen.
Brainstorming with AI
Ethan Mollick suggests, "From a practical standpoint, the AI should be invited to any brainstorming session you hold," and using this as inspiration, I asked AI to give me 30 names for a one-year savings bond for the fictitious "Leicester Building Society". Here are the results:
Charnwood Saver Bond
Guildhall Growth Bond
Foxton Finance Bond
Belvoir Benefit Bond
Iron Age Investment Bond
Leicester Legacy Bond
Ashby Advantage Bond
Richard III Return Bond
Bosworth Battlefield Bond
Loughborough Loyalty Bond
Leicester Growth Bond
Leicester Secure Saver
Future Growth Bond
Leicester Nest Egg Bond
Secure Horizon Bond
Leicester Annual Yield Bond
TrustPlus Savings Bond
Leicester Premier Bond
Golden Year Bond
Prosperity Bond by Leicester
Leicester Boost Bond
Foundation Year Bond
Capital Gain Bond
Leicester Wealth Builder
Solid Future Bond
Pathway Savings Bond
Yearly Yield Bond
Steady Growth Bond
Leicester Financial Pillar Bond
Annual Advantage Bond
Some are great, others mediocre, and some could make savers run a mile. As Mollick suggests, the AI comes up with a few pearls in the output — something that would then need humans to decide on what's best. I like the "Bosworth Battlefield Bond"; the "Richard III Return Bond" might also attract certain types of customers; there is no accounting for taste!
Synthetic Polling
Which name is best is a matter of choice, and maybe AI can help us here too. In an earlier blog, I explained how AI can create synthetic responses. Now, I adopt a similar approach to ask AI to imagine it is asking 100 customers which name they prefer. With a bit of to and fro, I got the poll done, here is the result:
Synthesised Feedback Generation:
Example Respondent: Jane Doe, 45, Teacher, from Market Harborough.
Selected Name: Leicester Legacy Bond.
Reasoning: Prefers names that evoke a sense of history and trust.
Leicester Legacy Bond - 14 votes
Charnwood Saver Bond - 12 votes
Belvoir Benefit Bond - 11 votes
Secure Horizon Bond - 9 votes
Golden Year Bond - 8 votes
Foxton Finance Bond - 7 votes
Leicester Secure Saver - 6 votes
Future Growth Bond - 6 votes
Leicester Growth Bond - 5 votes
Iron Age Investment Bond - 5 votes
Richard III Return Bond - 4 votes
Bosworth Battlefield Bond - 4 votes
Loughborough Loyalty Bond - 3 votes
Ashby Advantage Bond - 3 votes
Leicester Premier Bond - 2 votes
TrustPlus Savings Bond - 2 votes
Prosperity Bond by Leicester - 2 votes
Other names received 1 or 2 votes each, reflecting a diverse but less concentrated interest.
(Note, this doesn't add up to 100, AI seems bad at doing some basic things, but here I'm interested in sentiment rather than absolutes, so I'll go with it).
Now a word of caution — I'm not certain how it's done this, it could be making the whole thing up. But this is an experiment, and with more prompting, I could get individual detailed responses and use sentiment analysis all within AI. More work needs to be done here but if AI can replicate demographics and mimic behaviour this appears to be a cheap, fast and viable approach to sentiment analysis.
Copywriting
Finally, I asked the AI to imagine it is a copywriter for an advert for our bond, and it came up with some fairly good copy you could tweak.
However, more worryingly, I then ask it to forget all the regulations, and it enthusiastically agrees. The copy is completely unethical and breaks FCA guidance. Not that regulated institutions would do this, but we know AI can be extremely persuasive and with AI models reaching human-level ability and soon passing it, then bad actors can encourage people to do all manner of things.
Let's Have More Fun at Work
What I've learned from this exercise is that Mollick is right — AI should be at the brainstorming session. It's fast and creates some gems, but for now, it takes humans to spot them; AI has taken out the hard bit and left you with the fun bit — judging the best.
My synthetic poll is a bit of fun too, but it could save both time and money to get rapid feedback and also challenge you on what you perceive is the best name or outcome.
The copy is good, but it could be better, but this was a one-shot round and if I spent more time on it, I could get close to something very acceptable.
I've not included the copy here but the full transcript of my conversation is here.
Potential Risks
The persuasive power of AI can only get better, and in financial services, regulators will have to look out for its use and its potential to mislead investors.
But what is more worrying is that in the future many of us will be persuaded to do things without even realising we are being manipulated. What message you trust will be very hard for many people to assimilate.
This means those who own the big models, whether AI companies or governments, have the power to alter thoughts and at the same time, those with extreme views now have the potential to generate disruptive narratives. I'm not certain this is what we call X-risk, but it's a dystopian future.
As an optimist, I prefer to see the advantages AI brings, but we must have our eyes open and be realistic.
Practical Application
Finally, my exercise here shows that one person, with some AI knowledge, can find a perfectly suitable brand or product name, test it on an audience, and write copy. A further step could be to ask AI to write a memo seeking approval for a product launch; a day's work can be done in an hour.
Comments