Oppenheimer understood the dilemma. The science says it is possible. But what about the ethics of the decision to use that science and its potentially unforeseen consequences? He spent the rest of his life post Hiroshima and Nagasaki considering whether he had made the right choice.
Digital marketers do not face anything like the same degree of danger from their decisions. Yet there is a genuine possibility that what happens in the realm of social media analytics could either significantly enrich the quality of digital marketing - or potentially destroy it completely.
Predicting personality traits from social media data is a core objective for many digital marketers. Open source data provides the raw materials to understand individuals who are otherwise anonymous online - harnessing that power to target ads more effectively should move digital campaigns closer to long dreamed-of individualised marketing.
A recent research project has proven just how predictive this data can be. By analysing an average of 170 Facebook Likes provided by over 58,000 American consumer volunteers, two Cambridge-based researchers achieved accuracy of between 0.6 and 0.95 for significant - and often highly sensitive - dimensions of personality.
Their paper gained a lot of media attention because of the inevitable “Big Brother” comparisons. After all, being able to forecast 88 per cent of the time that somebody is gay based on what they Like on Facebook could have potentially harmful consequences. As could predicting if they use drugs (70 per cent accuracy), or what their race and religion is.
The authors recognise this political sensitivity themselves, noting that, “the predictability of individual attributes from digital records of behaviour may have considerable negative implications, because it can easily be applied to large numbers of people without obtaining their individual consent and without them noticing.” Their hope is that “trust and goodwill...can be maintained” through appropriate controls and transparency.
For the study, the research pair had the advantage of a data set to back-check their modelling of the open source Like data (although it is not entirely clear whether users of the myPersonality Facebook application knew they were helping a research project). In the commercial arena, it is unlikely that marketers would have such a large sample directly linkable to open source information.
It is also clearly the case that marketers tend to steer clear of the most sensitive dimensions which this study examined. Politics, race and sexuality have long been protected under the Data Protection Directive and it is a rare marketing campaign that can justify using these variables.
As digital data expands its reach beyond just shipping tins of beans, however, many of the attributes which the researchers looked at are starting to come into play. Insurers are very interested in whether people smoke or take illegal drugs, especially if they suspect that an applicant has not been fully truthful. Similarly, relationship status has clear uses by financial services providers (and others).
So the (data) genie is out of the (social network) bottle, just as the science of relativity led to the atomic bomb. This is precisely what worries the regulators in Brussels who are trying to ensure consumers are protected from automated decision-making that uses these types of data.
Yet it is worth asking whether these type of predictions are any different from the profiling currently in use based on conventional data. (This is also being tackled by those regulators, of course.) At a recent presentation by CACI which used profiles from its Ocean database, it was clear that the socio-demographic traits on offer were probabilities based on analysis of large numbers of variables.
That is exactly what the researchers provided - predictions, rather than certainties. It will never be possible to know for certain key variables for every consumer, leaving marketers to use whatever data tools they can to infer them. There is a danger of going to far - but for the moment, finding a consumer who has been harmed through the use of such tools is nigh-on impossible. As far as it goes, that is reassuring.