Prompting Unmet Needs

A Service Design thought leader wrote, perhaps with the help of a next word predictor, ‘Imagine how AI can help discover unmet needs.’

It struck me as odd to think that a pattern recognising system would be able to discover anything new, especially new and needful. Precisely because these machine learning systems connect things that are, for us, meaningfully connected, but on the basis of meaningless correlations seemingly imperceptible to us sensible social people, they often make mistakes that are not completely incorrect and so can be sometimes ‘interesting,’ meaning novel, but rarely significant. If there is anything significant in the output of a generative AI system, it is significance that you build into or out of what is generated. Reasonable (ignoring the theft behind the training corpus and the exploitative digital labour cleaning up correlations) uses of AI treat it only as a fuzzy front end starting point. So it strikes me as a misunderstanding of the nature of LLMs for instance to attribute to their outputs a capacity to ‘discover unmet needs.’

That phrase is itself is odd. It presumes that needs are unproblematic, perhaps essential. They are nevertheless things that can go unmet. To say 'discover unmet needs' is to claim that society appears to know that people have needs but chooses not to respond to those people’s needs. The excuse the phrase then offers is that some needs, or some people’s needs are, despite being unproblematic needs, hidden, and require discovery. There are unmet needs are out there.

An n-gram suggests that the phrase is relatively recent - of mid-century (Western) manufacture. The texts in which it occurs in the 1940s-1970s are mostly policy documents specifying empirical research into communities in order to brief different aspects of welfare state: ‘discover whose health, housing, education, etc, needs are going unmet.’

From the 1980s on, the texts containing the phrase are more marketing guides, and innovation manuals, and so by the 2010s, Design Thinking. Clearly the term ‘needs’ is now being used more loosely, covering things people can come to be convinced they need. To 'discover an unmet need' means to identify something that somebody does not currently feel is a need, but can be made to feel is needed, by the new device or device feature invented to meet that so-called ‘need.’ (I am reminded that the calssical rhetorical technique of inventio meant ‘to find’ rather than to fabricate something that did not previously exist.)

The issue is not these are ‘false needs,’ things that people do not need but can be made to want, strongly. More important to my mind is the ‘discover’ part of the formula. You apparently need to search for these ‘insights’ because people do not know what they really need. It is no use just asking them what they need, what needs they have that society is not helping with. In the condescending quip attributed to Ogilvy, “Consumers don't think how they feel. They don't say what they think and they don't do what they say.” People, this ideology insists, do not know themselves or understand their situation; whatever they declare as an unmet need is incorrect. You must have techniques for discovering their unmet needs precisely because ‘they’ cannot be trusted. Only data-driven, evidence-based systems can reveal to us, and 'them,' what their true needs are.

So when a service designer suggests that AI can, or even can help or help to begin to, ‘discover unmet needs,’ this seems to me just the latest way of disempowering the capacity of people to speak for themselves about what they feel they need. This is quite the reverse of what ‘human-centred design’ at its best (which it never intended to be so long as ‘(profitable) viability’ was given equal weighting with ‘desirability’) hoped to be.