St Mary's Academic on BBC Radio 4 and BBC World News

Dr Yasemin J. Erden, Philosophy Programme Director and Lecturer at St Mary’s University College, Twickenham was recently interviewed for BBC Radio Four and BBC World News focusing on the topic of CAPTCHA (Completely Automated Public Turing Test to Tell Computers and Humans Apart). Speaking on the Today programme for BBC Radio 4, as well as the Business Report for BBC World News, Dr Erden discussed CAPTCHA technologies, by which websites can differentiate between humans and spambots. They do this by generating and then assessing simple tests. These are designed to be relatively easy for humans to pass, while difficult for many computer programs. Typically this relies on the use of distorted text and unusual fonts and layouts. There has been renewed interest in this technology by the media, after a new software company claimed that their algorithms could ‘reliably solve modern CAPTCHAs, including Google’s reCAPTCHA’, with surprisingly high success rates of ‘up to 90 per cent on modern CAPTCHAs from Google, Yahoo, Paypal, Captcha.com and others’. [1] Yasemin explained to the BBC correspondents that even if these levels of success had been achieved (and it is difficult to assess this because the company so far has refused to disclose full details of the algorithm and information pertaining to these results), this does not mean that there are not alternative methods for CAPTCHA tests available. These include pattern recognition such as ‘put the food products into the shopping basket’; trivia like ‘what colour is grass?’; or very simple mathematical puzzles. In addition to this, some far stronger claims were made as a result of these apparent developments, including that they had created technology that could pass the Turing test [2] (a machine’s ability to exhibit intelligent behaviour), or that would lead to greater Artificial Intelligence (AI) capabilities. To this Yasemin responded that first of all, the CAPTCHA technologies are rightly considered ‘reverse’ Turing tests, since the original test was supposed to assess whether a machine could fool a human that it was also human (or in fact, that it was female), whereas the goal of CAPTCHA technologies is to allow a computer to distinguish between humans and machines. It is in fact far easier to fool a machine in these respects than it is to fool a human. Secondly, there has been no evidence given as to the way in which successes such as these (if indeed there has been success) would lead to any claims for better AI systems broadly speaking. Until there is further evidence for this (including peer-reviewed research) then there is as yet little reason to accept such claims. The interviews are available to view here.