The one-day meeting in Basel on “Using AI to Power Insights and the Business” had been much anticipated. It allowed us to deep-dive into the topic of AI in a more intimate setting following the discussions from the conference in June. AI in drug development seems almost like news from yesterday, but AI in business insight so far had less headline space. Can AI bring more than efficiencies? Can AI even help to find your killer insight?
Here is groupH’s pick of the key messages.
AI, for now, is “just another ‘person’ at the table”
To cut a long story short: there is much potential for AI as an assistant for specific uses but, generally, while it still has its flaws, the majority of presenters strike a somewhat cautionary note not to over rely on AI at this point.
The fact that AI (only) mimics human behaviour, is being trained for plausibility and not for truth, occasionally makes things up and is unable to explain exactly how it functions and where its information is coming from does not make a case for giving AI a senior role at the table. It would not for any human with the same traits.
AI is not yet ready – the audience is not yet ready
There is widespread experimentation with AI. In the main, individuals are using their personal versus their company logins and industry can only count a small number of experts. Perhaps the phase of ‘shock’ has passed, but before AI is being accepted as the new normal there is a phase of ongoing reorientation.
In Social Media Listening (SML), for example, AI is being trained to be more subtle and nuanced. However, it is not yet ready to frame a narrative within the context of research, opportunities and threats as Stephan Lebrat, Global MR Director at Takeda points out. And English language models currently dominate over other languages.
DALL-E 3 has just been made available by OpenAI as a generative visual art platform that integrates with ChatGPT and is meant to understand context better than its predecessor. https://openai.com/dall-e-3
A glimpse into the future of AI for agencies and industry?
Current AI language models make it possible to create artificial respondents whose responses to questions are difficult to keep apart from real respondents as Gemma McConnell from Day One Strategy explains. These AI respondents may be able to help with gaining an initial understanding of healthcare environments across therapy areas and indications at a much lower cost and much faster than undertaking PMR. This may be particularly helpful in conducting mock interviews preceding patient research; it may be particularly useful in rare diseases, where PMR is generally more challenging. At the moment, however, ChatGPT patient responses have a more generic feel to them and it is not known on which data they are based.
AI assisted desk research and first pass analysis of text, recordings and images helps agencies to become more efficient. An example of how AI is being used today for coding of data (e.g. for open ended questions) is the text analysis tool Caplena, https://caplena.com/en/. Researcher time spent with coding has been reduced by 46 – 67% according to Stephen Potts from Purdie Pascoe in a case study describing the adoption and implementation process.
Industry may be able to find a new purpose for existing data sets that support internal hypotheses generation but wants to make sure that its data is not training external models.
The personality profile of bots is typically highly agreeable and low in neuroticism as Paula Coyle from Blueprint Partnership explains. This may have an application in QC of demand forecast survey responses. In a self-funded study, the Mini-IPIP was used as a validated tool to segment physicians into under-, over- and accurate-estimators of their own prescribing. In the future it is thinkable that AI combined with behavioural science tools may allow to offer a more elegant and accurate calibration approach compared to traditional blanket calibration.
Fenna Gloggner from Idorsia Pharmaceuticals explained how Idorsia built its own internal ChatGPT model, an example also described by Ana Maria Aguirre Arteta, Global Governance Director at Novartis.
What we won’t forget from the conference
- Can AI bring currently more than efficiencies to business insight? No. Not really. But in large, data-heavy studies AI may become routinely used in the mid- to long-term, under the watchful eye of human intelligence (HI).
- Can AI help to find your killer insight? No. Not directly. But it may support making hypothesis and help analysis through consistent coding of transcripts.
- Gemma McConnell’s synthetic patient animation – close to indistinguishable from the real patient. Offering the mining of existing data repositories to supplement research with real patients.
- How will industry benefit from AI? Presented case studies describing efficiency gains far outweigh those hinting at insight quality gains from AI. Overall project value delivered by an agency remains driven mainly by HI for the foreseeable future.
- It may be too early at this point to count on efficiencies offered by AI before the technology is fully trusted by all stakeholders, However, over time these efficiencies are likely to materialise and may spur some competition among agencies to pass on some of these efficiency gains.
- The sigh of relief that ~80% of doctors were found being able to accurately estimate their own level of prescribing for a known product for the week after the question was asked (!) as Paula Coyle from Blueprint Partnership explained.