Why big business initiatives failed spectacularly and why exaggerated expectations lead to losses in the billions.
“The voice of reason is soft.” (Siegmund Freud)
The old world is dying, the new has not yet been born: it is the time of the monsters. (Antonio Gramsci 1937)
«Overpromised and underdelivered»: The ailments of Dr. Watson
In 2016, IBM had jokingly proposed the company's own artificial intelligence software Watson as US president (“Watson for President”).
A few years earlier, Watson had already won by a clear margin against two outstanding players in a Jeopardy game. This was hailed as a breakthrough in artificial intelligence. Groundbreaking applications in the legal and medical fields ("Dr. Watson”) seemed within reach and the business potential huge. The announcement of the MD Anderson cancer treatment center at the University of Texas that it would “eradicate cancer with IBM Watson Cancer” was to become reality, a gigantic market was to open up.
IBM invested heavily: In Munich alone, two impressive office towers were built for 200 million US dollars, the Watson IoT Center, designed for 1,000 highly qualified employees.
As early as 2017, it became clear that expectations had been set too high. MD Anderson and other hospitals terminated their cooperation. Despite the huge investments and huge amounts of scanned patient files, IBM Watson AI could only deliver what had been promised to a limited extent.
Even worse, the suspicion of misdiagnosis arose. The thought that Dr. Watson, to a certain extent with gauded physician title, as an AI system that is inscrutable to outsiders, is not or only with difficulty comprehensible to them, is a horror scenario, not only for the currently affected patients, but also for the treating physicians* (as programmed incapacitation of their competence contrary to the Hippocratic oath) and the clinics (anyway and in the USA especially because of threatening compensation claims under private law in enormous amounts).
Layoffs and failed projects at IBM Watson were the result. It is, said one affected person, as if IBM had big shoes without knowing how to walk in them. (A fate that by the way has been obvious for quite some time in the neighboring blockchain hype).
Standing against these developments, IBM's CEO Virginia Rometty stubbornly announced Watson's Law at the beginning of 2018: Following the already proven Moore's Law (doubling the computing power of microprocessors every two years) and Metcalfe's Law (the value of a communication network increases with the square of the number of participants), Watson's Law was to provide artificial intelligence to industry, consumers, cities, and simply to provide the whole of humanity with an independent and exponential growth compact that goes far beyond mere optimization.
Watson was to provide all mankind with an exponential growth compact.
Only half a year later there was no more talk of it. The AI strategy had not worked. The IBM management had made a mistake. One of the reasons given was the lack of data quality: Dr. Watson had been fed with all available medical records for a long time, without any deeper reference to their quality and the respective context. The currently propagated AI, however, is as weak AI only man-derived, syntactically formalized intelligence for a specific area. It can only recognize patterns and correlations, but cannot understand causalities. AI can therefore not develop itself, it is not able to learn by itself.
AI is not capable of learning on its own.
Prejudices and errors are algorithmically petrified by such AI. Data are useless for a computer if they are not properly “labelled” by humans and stored with a concretely thought-out model, against the background of which they can be evaluated purposefully as information. Once there is too much blur (entropy) in the data, every algorithm fails.
After painful losses, IBM cut the Watson team in half and finally bought it in October 2018 from Red Hat, a solid provider of Linux solutions, with its multicloud solution far less glamorous than the promises of artificial intelligence.
After wasting time and money on its exaggerated AI expectations, IBM is now looking to catch up with its pre-print competitors such as Amazon, Google and Microsoft in cloud computing. However, they now have quite a head start.
Whether and how IBM will be able to make up for the losses from its overestimation of AI remains to be seen.
«Overhyped and too broad»: GE Predix as an industrial hut worth billions
In 2016, Forrester named Predix (predix.io) as one of the leading solutions in the field of industrial Internet.
GE Digital had full-blownly announced Predix the previous fall as a comprehensive cloud-based service platform for the industrial Internet of Things (Industry 4.0): *No less than 20% savings should be possible everywhere, across all industries.
In 2016 alone, GE Digital 6 billion US dollars has been invested in the specially created GE Digital 6 billion US dollars. GE wanted to generate 14 billion dollars with this in 2020, when one of the “top 10 software companies “ in the world, according to the then CEO of GE Digital, Jeffrey R. Immelt. Data, supposedly the new gold, was to be collected across all industries and then made accessible with the help of smart algorithms. The more and the faster, the better. But that didn't happen.
Overhyped and too broad – an epic fail for the ages.
Overhyped and too broad, that's how the approach is seen today, an epic fail for the ages. Employees have been laid off. GE Digital was taken out of Immelt's successor, after the originally full-bodied “Digital Transformation” had been announced, but the bottom line was that so much money was simply wasted that some observers were already understandably concerned about the survival of GE itself.
The mistakes were manifold: GE wanted to compete with Amazon, Google and Microsoft for the best cloud technology of its own (unpromising), on the one hand, and AI and digitization as a commodity, on the other hand, to be applied crosswise and generically to all industries (unrealistic). And GE believed - what a cardinal fallacy - that digitization was a dependent production factor from which the business would continue to develop independently.
Digitization and AI are not independent production factors but accessory ones.
But data is not information and AI systems don't understand anything. This requires industry know-how, i.e. people who train them. Modesty is a virtue and data can be refined, but not gold in itself. Often data must first be prepared and recorded in a targeted manner so that it can be used to create value. “Use of data requires [human] prediction”, as William Edwards Deming, a mathematician and statistics professor, already formulated it in 1993 [Supplement PE]. *Big data as fat data, however, obviously only devours hard money and valuable resources.
Accenture & Co as digital arsonists?
In 2016 Accenture published a press release: Germany is thanks to artificial intelligence facing a sustainable growth spurt. An in-house study proves that economic growth can be doubled by AI by 2035, and labour productivity can be increased by up to 40%. In cooperation with a market research institute, the potential for growth through AI was meticulously calculated for each country and explicitly reaffirmed for the Austrian Minister of Economics and Digitization.
With the help of AI, economic growth could be increased and thus the time to double the economic growth for each country could be shortened considerably. Without AI, for example, the German economy would only double in 50 years, whereas with AI it would already double in 25 years!
Regardless of whether one takes an undivided, positive view of economic growth or feels close to Greta Thunberg: When thinking of a quantitative doubling of the total global economic output in the next 25 years or even earlier, almost everyone will get worry lines in view of the climate problem. Who is going to buy all these products and services? And where to put the double amount of plastic waste? Both the naivety with which it is claimed that AI will make this possible and the naivety with which the growth paradigm that has become necessary is simply to be continued are breathtaking.
The naivety and audacity with which AI is praised as a panacea is breathtaking.
This is possible because AI not only optimizes, but is simply a fourth generic production factor next to labor, capital and productivity.
- The mayfly “Watsons Law” by IBM as well as the assumptions by GE, which have meanwhile been proven wrong, send their regards.
According to Accenture, AI systems can perceive, understand and act. This makes intelligent automation possible. This has the ability to solve problems across industries.
- GE's misconception of AI's industry diagnosticism sends its regards.
For example, a program called Amelia can read technical manuals and “diagnose a problem and suggest a solution” and thus instruct maintenance engineers worldwide.
- Greetings from Dr. Watson, who wanted to provide this almost verbatim support to doctors who were exposing their patients to undefined risks.
Amelia has also learned the answers to the 120 most frequently asked questions to mortgage brokers by heart and can answer financing questions, which naturally saves a lot of work.
Now all that remains is to inform the customers which questions they are allowed to ask and which ones they are not allowed to ask under any circumstances, then a lot of work can certainly be saved.
But that's not all, Amelia can identify and close gaps in her knowledge like a duty-conscious employee. For this purpose, the program delegates difficult questions, the answers to which it does not know, to a person and remembers their answers. This self-learning aspect of Amelia is a fundamental change. At the latest here the grounding is lost:
According to serious AI research, which incidentally also exists, self-learning AI is still a long way off (see below: Ford) or will not be feasible for reasons of principle, at least not in the naïve form envisaged here.
Capital assets would thus automatically increase in value. AI as the digital perpetuum mobile of capital accumulation!
Optioment and other scams: Bitcoins as the black hole of human intelligence
Memories of the almost analogous promise of Optioment appear.
Optioment was one of the many Bitcoin and Blockchain scams, a fraudulent investment scam. Optioment promised exactly the same, namely automatic asset growth, only in this case with block chains and not with AI. As a fraudulent snowball system (Ponzi scheme) in Austria, it caused a circulated damage sum of 100 million Euros for over 10,000 affected persons. Optioment was also founded in 2016. Of course, the Optioment case cannot be compared with the failed projects of IBM or GE. In all three cases, considerable sums of money were burned in the wrong belief in the promises of digital technology.
In 2016, misconceived exaggerated expectations of digitization caused billions of dollars of damage at IBM, GE and Optioment alone.
The increase of capital and labour is, again after Accenture, the second promising area of application. Dissemination of innovations the third. Autonomous vehicles would reduce the number of accidents and thus possibly represent the largest health initiative in the world.
To counter the disadvantages and help people who become unemployed through AI, “redistribution effects” should be set. Indications of positive climate effects and broader access to good health care could be used to counteract the mood and sell these false digital promises politically.
The ICT sector (including cloud computing and thus also AI applications) is already estimated to account for almost twice as much greenhouse gas emissions as civil aviation worldwide.
According to the study, employees could in future be much more involved in creative tasks and innovation issues that would result in higher added value than the jobs they have done up to now. For example, productivity in agriculture could be increased by up to 38% through the use of self-propelled tractors and milking robots that “taste” the milk at the same time.
It is not clear how much digitalisation will increase productivity. So far, the expected effects have remained well below expectations. Some economists like Robert Gordon doubt whether digitalisation can increase productivity at all.
The Future Council of Bavarian Industry was surprised to note that “no correlation between the degree of digitisation and the productivity growth of the national economies can be recognized”. It justified this – a comprehensibly undesirable insight in a study titled “New value creation through digitisation” (PDF Download, page 38) – with the poor data situation.
Economists doubt whether digitization can increase productivity at all.
And human needs remain essentially the same despite digitization, at least they are not increasing exponentially. In any case, the efficiency of individual tasks will increase dramatically and people will no longer be able to work in their traditional areas. Therefore, clear framework conditions and accompanying adjustments in technical, legal, organizational and social terms are necessary. A simple continuation of the current work organisation with AI support will not work, accompanying measures are necessary.
According to Accenture, however, the data, the new gold, would have to be released from the “data vaults” of the offices. “The use of artificial intelligence requires the most unhindered access possible to data. The culture of official secrecy prevents this. (emphasis PE).
For private commercial interests, public data should be made available free of charge under false pretences.
For similarly as IBM “freed” Watson from their registers, medical records which are now electronically accessible but have not - as hoped - become self-learning. China can also serve as a dystopian example. There, unhindered access to the data of all citizens is possible. A social score is used to categorize people and regulate their freedom. And yet the growth rate in China has been falling for some time now.
Accenture, a consulting group with more than 450,000 employees worldwide, is pushing to break open public data vaults for AI projects to achieve goals that IBM and GE have used the same arguments to undermine.
It would be tragic if data protection, individual privacy and personal rights were actually undermined because of such transparent, misleading and implausible arguments.
Even official secrecy, which is a smoke grenade for political arrogance that is democratically dubious, suddenly receives an unexpected justification as protection of the common good against unauthorized access for profit.
Emerging technologies and digitalisation will undoubtedly bring us improvements and competitive advantages. However, the right framework conditions must be set in ethical, social, organizational, technical and legal terms in order to achieve sustainable value creation.
Digitisation is not an end in itself and certainly not a self-evident goal. In order to implement it successfully, it requires goals and strategies - and above all expertise.
Digitisation is not an end in itself and requires human expertise.
This common sense often seems to be lacking in the area of public administration (as is the case with the Austrian Ministry of Digitisation, for example, with the registration app – which did not check registration addresses, the election app – which did not forward applications for voting cards, and the supplementary register – which placed uncontrolled personal data on the Internet).
The time-delayed bouncing back of untenable promises of redemption will not get us any further. It is shocking to see the naivety with which politics stubbornly continues to cling to old paradigms such as economic growth and weakens basic legal values.
As the examples of the American industrial legends IBM and GE have shown, however, smoke grenades and theological-seeming promises of salvation will not last. In the case of these corporations, it did not even take three years for the false assumptions to end in losses running into billions and tough corrective manoeuvres.
Since artificial intelligence will inevitably always be derived and orchestrated from human intelligence, it will be possible to meet existing needs more cheaply and more quickly. Although it can increase productivity, it does not represent an independent production factor.
Artificial intelligence is not an independent production factor.
The management problem we actually face is a different one: the ever-increasing complexity of our world (Stafford Beer, Fredmund Malik). AI can only reduce this complexity to a limited extent, because it is always under-complex to the human one. *However, an under-complex system cannot control a highly complex one (Ashby's Law of Cybernetics), nor can it with a lack of technical penetration of a government policy.
Moreover, digitisation will fundamentally change the internal structure and organisation of almost all companies and the administration. Existing processes will be automated 1:1, rather than simply being orchestrated digitally from the customer's point of view. The internal conversions must be accompanied accordingly with foresight, employees must be integrated in a different way and in a new way. This requires new forms of cooperation and work organization (keyword New Work, Deming – The New Economy). IT security must also be rethought against this background, there is a real need for action.
Sub-complex introduced digitization only leads to a dead end.
What AI seems to have ahead of us in terms of speed and precision, it lacks social competence, the ability to interpret content, situational presence and the ability to anticipate developments. Content-related know-how, careful analysis of data, the design of the required information structures and the recognition of causalities and system connections is required, no naive data digitalism as belief in unfulfillable prophecies of sales-strong consulting firms.
In this way, AI can help us to solve our problems. But it cannot do this for us, even with billions of dollars of investment up front.
The risk is high, the possible damage to defunct consulting fees, eroded data sets, eroded legal foundations and the loss of traditional core values is high. Government digitisation initiatives deserve more goal orientation, clarity, sustainability and plain common sense.
It obviously needs more natural intelligence to use artificial ones correctly.
About the author
Dr. Peter Ebenhoch is multidisciplinary business agilist. As a certified strategy (HSG) project (PMP) and risk manager (PECB), he supports organizations and companies with regard to sustainable digitization and disruptive intelligence.
- Watson for President 2016
- Watson (Künstliche Intelligenz) bei Wikipedia
- Strickland, Eliza (2018): Layoffs at Watson Health Reveal IBM’s Problem With AI. Axed engineers say IBM isn’t always smart about artificial intelligence, IEEE Spectrum
- Strickland, Eliza (2019): How IBM Watson Overpromised and Underdelivered on AI Health Care
- Woodburn, Doug (2018): ‘Watson's Law’ will rival Moore's Law, claims Ginni Rometty in AI keynote, CRN Channelweb
- Scofield, Jack (2019): Can Red Hat save IBM?, ZDNet
- Bridgwater, Adrian (2018): IBM CEO Rometty Proposes ‘Watson's Law’: AI In Everything, Forbes
- Herper, Matthew (2017): MD Anderson Benches IBM Watson In Setback For Artificial Intelligence In Medicine, Forbes
- Chen, Angela (2018): IBM’s Watson gave unsafe recommendations for treating cancer, The Verge
Ford, Martin: Architects of Intelligence, The truth about artificial intelligence, Packt Publishing, Birmingham 2018.
Gordon, Robert: The Rise and Fall of American Growth
Zukunftsrat der Bayerischen Wirtschaft (2017): Neue Wertschöpfung durch Digitalisierung Analyse und Handlungsempfehlungen, PDF Download, «Es ist […] kein Zusammenhang zwischen dem Digitalisierungsgrad und dem Produktivitätswachstum der Volkswirtschaften zu erkennen» (Seite 38)