Mind the Gaps: A Critical View on PwC’s AI predictions

New year, new predictions, new consultancy hours to be sold. PwC starts their “Sizing the prize” analysis with a claim of worldwide GDP increase of $15.7 trillion within year 2030. We have previously seen claims like these with new technologies, most famously with the productivity paradox in the 1970’s and 1980’s with the adoption if IT technologies where “As with previous technologies, an extremely large number of initial cutting-edge investments in IT were counterproductive and over-optimistic”. I fear that PwC’s report is painting a similar picture, especially as you start to dive into the second part of the report. Here PwC presents the AI impact index of which industries are most ripe for disruption, and it is here the naive and simplistic assumptions starts to shine through.

The AI impact index

Overall the ranking is what we are used to seeing with healthcare, automotive and financial services as the top-ranking industries for AI disruption and AI gains. Here is why I would be much more sceptic.

Healthcare

Here PwC briefly mentions the barrier to overcome for AI to make gains as the complexity of human biology, and the ability to use a patient’s unique history as a baseline for flagging possible health conditions.

Now the complexity of human biology happens to live in a analog medium, with a define timeline of progression of different changes and reactions. No amount of digital innovation is going to speed up that process, which ultimately means we have to conduct time consuming experiments, and observational studies to get empirical evidence to support, train and validate whatever we want to do with AI. Furthermore, anyone who have ever worked with translational medicine knows just how poorly assays, in-vitro work and even animal models translate to human clinical studies. For AI to really make a dent here, we need more fundamental biological research and understanding.

As for the second part on patient’s baseline measurements. Yes I agree, the problem is that we don’t have those. Think about the economical impact it would have on healthcare if we were to establish healthy baselines for an entire population. Moreover, research have pointed to the fact that intra-individual variation is often larger than inter-individual variations, meaning we would need multiple longitudinal measurements to establish the baseline range of measurements. While I am quantified self person on many levels, I do not get regular blood work done, or cytokine profile measurements. It is simply too expensive, and the cost of this at population scale would require a complete change of our healthcare system at the promise of some unrealised future gain in overall decrease in mortality and disease rate?

Automotive

Autonomous fleets, just around the corner. In 2016 Elon Musk claimed that their system was likely better than human drivers, in two to three years – so 2019. That is 6 years ago it should have surpassed human driving capability. In 2019, the claim is that mid-2020 over a million robo-taxis will handle transportation. While the field is steadily improving and we are seeing both level 3 and level 4 autonomous vehicles in select areas, I would be hard pressed to see that the next 5 years will bring fully autonomous fleets operating 24/7 to be a standard. And there is the small problem of consumer trust and regulatory approval. It only takes a few 18-wheeler accidents to completely shut something like this down. With the margin of error being extremely low, and the effects of any error likely to impact all autonomous systems I see this as a much longer game.

Financial services

While PwC might be correct that personalized financial planning can be theoretically done using AI, it does not account for the underlying structural problems in the economy. Here I am thinking about the increasing wealth-gaps, the differences in regulation between private and institutional investment, access to market on identical terms, and the fact that 50% of the US population is currently in credit card debt. You need to have money to invest before AI will be an equalizer within financial services. Add to that regulatory mechanisms that are now being enforced to limit leveraging, debt taking, taxes on unrealized gains etc. AI will not make these structural barrier disappear, but AI will likely be a game changer for institutions, high-frequency traders, and similar which are already at the top of game. Likely causing more concentration of wealth.

Retail

While not in the top ranking, PwC is enthusiastic and perhaps a bit naive about the future of personalized design and production. While I do agree that consumers will spend “less time exploring shelves, catalogues and websites” based on better recommendation and consumer understanding it will take time. Mainly because the current system works, and there might be faster ways to grow revenue than finding completely new ways of understanding consumer wants and needs. Their second argument on fully personalized products as “made in small batches using automated production” I don’t buy. This is completely ignoring the fact that this requires AI and robotics driven manufacturing to become a commodity, and it ignores the fact that multiple consumer items needs various regulatory approvals from CE marking in the EU, to global and differing safety testing for cosmetics. This often at the final product stage, making bespoke manufacturing at scale problematic.

Realizing the potential

Here the PwC report goes into how to make strategic decisions of whether or not AI is something your industry should worry about, and how you are to approach it. Here is their biggest blunder – a technology first approach: “how different AI options help deliver your business goals”. First, it treats AI as a smorgasbord where you pick and choose your accelerator, and it aligns it directly to business goals – but the business goals you can achieve with AI in your industry might require you to scrap your decade long focus and cannibalize your current market cap to stay relevant. Hence you need to stay true to what your business vision and mission is, then contemplate if AI will fundamentally change what your are doing, and then start to look into the actual technology.

Take for instance a security camera manufacturer. Today the goal is to sell hardware, but there is a limited market, and it is pretty saturated. To find new revenue streams you can pivot, using AI to do threat detection on your cameras and become more of a vertically integrated service provider. Now this follows a different business model which could include subscriptions, different levels of detection, continuous maintenance of your security system (both hardware and service) which could have a larger market, since you are now offering something your competitors do not. This aligns with a company vision of delivering safety and security, but it does not align with a company vision of making the best cameras.

My Perspective

The PwC report reflects a common consulting industry pattern – creating urgency to drive advisory services. However, it has its merits. The report offers some key ideas around workforce development, the growing importance of soft skills, and the societal implications of AI adoption. I would have liked these ideas to be explored more, rather than focusing on the oversimplified industry impact analysis that dominated the report.

The fundamental flaw in PwC’s approach lies in its assessment of AI’s industry impact without any deeper consideration of industry-specific complexities, regulatory frameworks, and the potential need to reimagine current business models.

Yet, the future of AI is in my mind promising. Its true potential is not to make current business models more efficiently, but in fundamentally reshaping how we address human needs, enhance decision-making, and remove repetitive tasks.