What’s new
OpenAI – the company behind ChatGPT – has announced a series of models that “think more before answering”. This set of models is called OpenAI o1 models.
This is a big step forward for AI. The key difference is this: under the old paradigm, large language models (LLMs) like ChatGPT could really just think in one direction. They essentially come up with one word, and then figure out what they next word is most likely to be – and then keep doing that. A little like how it feels when I am answering a question on the spot about something that is new to me.
Deep thought
The difference with these new “o1” models is that they can essentially go away and think about something. More like when I am asked a question by email about something that is new to me. I will start by running through a chain of thought – like the “next word prediction” model above. But then I take that answer and try and think about whether it is actually correct – taking that answer and starting a new chain of thought. I will also likely come up with other trains of thought from scratch to give me other possible answers. Then I will start new trains of thought to compare the possible answers to work out which one makes most sense. This is what the new o1 models do.
These new models don’t give you the first answer that comes to their heads. They come up with a bunch of possible answers. They test if those answers make sense. They compare their answers. They make improvements.
The cost of this is computing time – and so money. The same as if I need to answer a question on the spot about something new versus going away and thinking about it to give a considered answer. On the spot this might take me a minute. Going away and thinking about it might take me hours.
But the benefit is that it gives us much smarter answers.
The advertising business
Now coming to how this affects advertising. OpenAI o1 and other “deep thinking” models that will inevitably be released by other companies – will be able to give us deep thought. For lovers of the Hitchhikers Guide to the Galaxy – it looks like this will be much more useful than Deep Thought there.
If we break down the processes in advertising, there are a lot of places where we have deep thinking (one or more people going away and figuring something out for a period of hours or days). This is all human at the moment. But AI will now be able to do more of this.
In the advertising process we have: market research -> strategy -> creative ideation -> creative development -> creative execution -> production -> media. For each of these steps, there is deep thinking to be done. AI models like ChatGPT are already being used extensively – partly as a replacement for Google search and partly to act as a colleague who can come up with a draft of something for you. The problem is that the answers are not always very good. It is still obvious when something is written by a model like ChatGPT or Claude – and what is written almost never makes complete sense. That said, it is reducing the time people need to spend on things by providing a first draft which they can then work on.
The o1-type deep thinking models will improve the quality of output we get from AI models. They will give us smarter answers.
No more humans?
The question then becomes what will happen. Will deep thinking AI models replace the need for humans in advertising? Will we no longer need creatives, client servicing teams, accounts teams, strategists, media professionals and production professionals? Will clients be able to directly tell a deep leaning AI model all about their brand and what they are trying to achieve through advertising for their business – and the AI model will be able to do all the thinking and work needed to go through all the steps of the advertising process: doing the market research, coming up with strategy, coming up with creative ideas, working out which ones will be most effective for the client, developing the creative, executing the creative, producing the final creative assets in whatever form, picking the most effective media, deploying the campaign to the right media, collecting performance data, and interpreting performance data?
By (deep) thinking through all of this – I think the answer to all of these questions is no.
AI is like Photoshop
Deep thinking AI will becoming increasingly important. But it will be a multiplier rather than a replacement. Advertising firms and clients that adopt AI early will have a major advantage until others catch up. But ultimately the impact on advertising will be like the introduction of creative software (like Adobe Illustrator and Photoshop) in advertising.
To look at this, it helps to break down advertising to it’s simplest model. Brands want to convey information and emotions to humans. There are a limited number of humans out there (even less in a brand’s target market). And each of those humans only has a certain number of minutes in every day during which they are exposed to advertising (this might increase or decrease with AI but is still limited). So brands still have to compete for those limited eyeballs and that limited human attention.
Because of this – media will still be limited and expensive. And because of that, brands that produce the most effective creative for their purpose will still achieve the best return on investment on their media spend. That means they will be the ones who are able to outbid their competitors for the still limited media available.
Raising the bar
Deep thinking AI models could in a few years do almost all of the steps in the advertising process. Even now, a brand could use deep thinking AI models to go through all the steps themselves. But having excellent humans working with these AI tools will produce much better work than the tools by themselves.
The analogy with creative software is helpful. When creative software became available, a brand could have taken Adobe Illustrator and produced creative assets for which they would have previously needed an advertising agency. But this did not happen for large brands – brands continued to use advertising agencies – and the advertising agency used the software. What did happen was that the bar for “good enough” advertising was raised.
A similar thing will happen with deep thinking AI models. Some tasks will become less important – for example data processing of survey results might need fewer people. Optimising media spend might need fewer people. Translating copy for local markets might need fewer people. But in each of these areas – the higher bar through competition between brands might need a greater of number of people for other tasks. For data processing of market research results – there might be a lot more work on collecting the right data and working out which data will have the greatest business impact for the brand. For media spend – time will be spent on adding more media types of the media mix that is optimised. And in localising advertising for international markets, competition will incentivise brands to spend more on modifying advertising more to better fit local cultures so that conversion rates are higher in that local market – to prevent being outcompeted for media once all brands can use AI models to correctly translate copy.
Why do we still need humans?
But why do we still need humans? Why can’t AI do all of this? The answer is that AI models are a different type of intelligence to humans. An analogy here is that being a Nobel-Prize winning physicist is a different kind of intelligence to being a top popular music songwriter. Deep thinking AI models are likely to quickly get exceptionally good at taking large amount of existing information, and summarising that. And they will be good at finding the right information to find and summarise for a particular problem. For example if you ask a deep thinking model to write a hit song about heartbreak suffered by a two-headed, three-armed, hedonistic, self-centred alien – and you want this poetry to win a Pulitzer Prize for Poetry – the model will be able to gather the poetry of previous prize winners, it will be able to get poetry and other descriptions of heartbreak by hedonistic and self-centred people. It will be able to get information about aliens. It will then be able to combine this and other relevant information to write a poem.
This is somewhat similar to what humans do. But two differences are that a human will be able to come up with new ideas at the edges – which at least for now is difficult for an AI model to do well; and the human will be able to tell us of experiences they have lived in their poem – which as human consumers we seem hardwired for. For example, with the advent of photography in the 18th century, the value of human artists to precisely paint what they saw fell. But artists moved from realism to abstract, emotional and conceptual art. This would have been difficult to predict. But now this is a large employer of human time – in everything from fine art to fashion to comics to architecture to video games. For the real-human-story attraction point, a good example is chess – a lot of humans (I expect more than ever) watch other humans playing chess on platforms like Twitch – even though computers have long been able to consistently beat the best humans at chess. We like the human aspect – the story behind the player, the emotions and perhaps some other things related to them being human. This seems to have evolved strongly within us over millions of years.
So for advertising, I strongly expect that we will see more people employed in this profession over the next ten years. Across pretty much all stages of the process.
The way forward for agencies
For agencies, the formula is to keep adapting to developments as they arise. Like snow boarding down a new run – you don’t know exactly what is coming up – but you know the general direction and keep course correcting very frequently. This means rapidly bringing in new technologies, without building long-run infrastructure around any single technology – as the state-of-the-art will continue to change rapidly. It means giving your staff access to AI tools, and having your staff use these to reduce costs and more importantly raise the quality bar for clients.
The way forward for brands
For brand, it means working with agencies who are most effectively adopting AI. The agencies who you can see are raising the standard of output they give you per dollar spent – by using AI to automate some of what was previously done by humans – and instead using that time to materially increase the quality of your creative work – as measured by its ultimate impact on your customers and the translation of that into your long term business goals. Other brands will inevitably catch up in quality – but you might get a two or three year head start – which could dramatically increase the long run competitive positioning and so value of your brand.
