What does AI have to say about the Trump vs Turnbull Leak?

Our weekend papers were filled with transcripts of the ‘robust’ conversation had between President Trump and Australian Prime Minister Turnbull on the issue of refugees. Some commentators were concerned about how ‘leaky’ the Trump administration had become. Others welcomed the transparency afforded by the transcripts, allowing the public to see beneath the public personas that many leaders are adept at maintaining. One thing is clear however, increased digitisation is providing us with at least the technical capability to track virtually everything anyone says to each other over digital channels. Stopping the leaks through purely governance mechanisms alone is like pushing the proverbial uphill. In the end we can only rely on developing the appropriate cultural and behavioural ethical norms.

Accepting that an increasing amount of the content we exchange with others is being captured digitally, how can technology use this for more good, than evil? Specifically, how can technologies like Artificial Intelligence (AI) engender more constructive and positive behaviours. At SWOOP we have been developing behavioural personas, based on the interaction patterns identified from Enterprise Social Networking (ESN) platforms like Microsoft’s Yammer and Facebook’s Workplace. We have drawn from research conducted by MIT on Social Physics and Google’s Project Aristotle to identify the interaction patterns that result in superior collaborative performances, especially within teams and groups. More recently, we have been adding Sentiment Analysis capabilities to enrichen the behavioural measures we report on.

Sentiment Analysis

Sentiment Analysis is a mature field for AI, with many vendors now providing such capabilities. While still not perfect, our customer trials using the Microsoft Cognitive Services Sentiment Analysis engine, together with a degree of our own customisations, have proved to be useful enough to formally install it in the SWOOP product. You can find out more about how SWOOP employs sentiment analysis here. The remainder of this post, however, is aimed at how we see an expanded use of sentiment analysis might help to achieve the positive and constructive behaviours we all seek, within our organisations. And what better place to start than the Trump vs Turnbull transcript.

Trump vs Turnbull: what AI can tell us

The traditional use of sentiment analysis by online community managers is to identify areas of online dialogue that might be seen as offensive or damaging to the community; so that appropriate interventions can be made before the damage is done. If for example, a discussion of the Trump-Turnbull ilk had occurred on the on a company’s ESN, how would the SWOOP sentiment analysis report on it? Well as the newspapers were kind enough to provide us with the full transcript, we were able to copy it into our own ESN to see what it came up with:

The SWOOP sentiment analysis module breaks each message into sentences and rates each sentence, courtesy of the Microsoft text analysis engine. On the left of the screen you can see an overview of the sentiment pattern, with green sentences showing positive sentiment, grey being neutral, or passive sentences, and red being negative sentiment. We should point out up front, that ‘negative’ does not always mean ‘damaging’. In fact, in context, a negatively scored sentence could be very positive in terms of constructive input e.g. “I think that the problem with that idea is that …” would be flagged as a negative sentence. Community managers can click on a sentence to expose the content, to make this judgement themselves. However, without even looking at the content, the pattern on the left hand side tells us a lot about the nature of the interaction:

As we venture further down the ‘sentiment scroll’ we see a pattern emerging:

The MIT social physics research tells us that productive interactions are characterised by short and frequent interactions. This little episode sees an imbalance of message size between the participants, so potentially not constructive. However, when we add sentiment, what appears to be a bullying activity on the part of Trump, sees Turnbull not react similarly, but to remain positive and succinct. The last Trump message shown, now has a mix of positive and negative. With the benefit of reading the content, it is in fact a begrudging acceptance by Trump that he would need to honour a deal made by the previous administration, but he didn’t have to like it. Overall, the commentators suggested that Turnbull came out looking pretty good.

So what can we expect for the future?

The initial MIT social physics experiments used social tags to collect non-linguistic interaction data, akin to the way SWOOP devises its behavioural personas today. By adding linguistic filters, we can develop an even richer picture of human interactions. The interaction patterns described above are only the tip of the iceberg, in terms of insights available through analysing digital interactions. A pattern of team interactions that show balanced interactions amongst team members, of not just message lengths, but also message sentiment, can predict stronger performance. A totally ‘green’ positive interaction pattern might flag a positive culture of shared admiration, but difficult problems are more likely to be solved by a balance of positive and negative interactions. A totally ‘red’ negative interaction pattern would be what one would expect, quite destructive.

Individual privacy must still be protected however. Therefore, we see for the most part, the patterning analysis of the type we describe here, could be made available at aggregated levels of a whole enterprise, a business unit, group or team, to assess the health of these groupings, and to eventually predict future performance. That said, we would also see value to the individual, in privacy protected self-reporting. For example, are cyber bullies really aware of their negative behaviour? Can the positive reinforcer also bring themselves to provide constructive criticism? The data on personal patterns of interaction can help with this.

Whether we like it or not, digital transparency is on the increase and here to stay. If we can harness technologies like AI to facilitate more constructive collaborative behaviours, then hopefully, we can anticipate that the world will be a better place for all.