Artificial intelligence not possible for data analysis with IBM-Watson2019-09-05T19:44:34+01:00
Service provider for data analytics and software development on Facebook Our Twitter page - software development service provider Data analytics and software development corporation at LinkedIn

IBM Watson fails – “AI” Watson

Because of Watson’s sluggishness IBM boss Virginia Rometty should no longer speak of “AI”.

Updated version of text *IBM Watson fails* 2018:

The following text deals with the failure of IBM Watson in the insurance industry, in particular in international insurance groups. The text was prepared in 2016 and 2017.

This being the case, we’ve been pursueing IBM projects in different industries over a longer period of time. We do this because a) we are increasingly coming into contact with projects that have failed with IBM Watson before and b) we are generally interested in AI progress anyway (not just IBM related). With regard to point a), we are, unfortunately, only contacted by corporate customers when Watson projects have been cancelled and when there is hardly any project budget left for the introduction of other machine learning solutions.

In 2018, we were confronted with a project in the US in which certain types of cancer were to be diagnosed by applying artificial intelligence that relied on medical imaging techniques. Watson,, thus, had already been fed millions of images from X-ray, CT and MRI machines, as well as with results from small synthetic drill holes. But Watson was still routinely misdiagnosing and recommending wrong and unsafe cancer treatments. The summary of the study is publicly available in the original English text on the following page:

At the moment, we at HighPots keep the opinion that Watson cannot, should not and must not replace doctors. We even go one step further and claim that you as a company will not get solutions faster with Watson than with the use of classic software (e.g. from SAS) or with your own developments (e.g. with the deep learning frameworks TensorFlow, Keras or Apache Spark). Should you require Deep Learning consulting or technical implementation in this regard, we look forward to hearing from you.

End of updated text *IBM Watson fails* 2018. Below you will find our text from last year.

IBM Watson is failing. More and more companies are reporting on the failure of IBM Watson. We as Software development service providers and Data Analytics service companies also had our own history with IBM Watson. Especially in the realization of a customer order in the area of KI-based Dynamic Pricing. But also in marketing automation, especially in dynamic real-time assignment of user profiles to target groups, we quickly reached our limits with IBM-Watson. The marketing department of IBM-Watson seems to be too fast for IBM-Watson product development.

The failure of IBM Watson – the story from the scratch

We first became aware of IBM’s miracle machine “Watson” at the Nobel Laureate Meeting in Lindau. During the break I heard IBM’s chief architect and quantum physicist at the next table say:

“When we connect our Watson AI to the quantum computer in the future, we will have an unbeatable B2B application. We are currently conducting initial experiments with quantum simulations in which we are identifying scenarios for the B2B application of Shor algorithms suitable for quantum computing. We are also looking for further mathematical possibilities in which quantum computers can play out their advantages for B2B business fields.”

IBM Watson and Tesla have one thing in common: Both companies are world champions in making announcements

In my blog post of August 5, 2017 “Quants solve bits ab” I already reported about IBM’s open innovation strategy and the possibility to program quantum computer simulations free of charge. At present, however, the expected hopes of the Open Innovation Strategy do not seem to be fulfilled, as new ideas regarding mathematical models for quantum computers are not forthcoming.

It seems that IBM has no choice but to hope for progress in the development of a quantum computer. In my opinion, the artificial neural network behind Watson is only suitable for a few application scenarios. To speak of “AI” here is almost cheeky; although Amazon’s Alexa is similarly weak in the B2B environment, IBM has at least a few years of development lead.

Insurances that assess risks better make more sales – but IBM Watson is no help

The handling of large amounts of data has always been part of the business model of insurance companies. Anyone who correctly assesses the risks of this world on the basis of high-quality data can earn considerably more.

Tender period and the smart IBM Watson sales

When we received the invitation to tender from the Sompo insurance group via the Japanese Chamber of Foreign Trade in 2014, we analysed it carefully. Sompo wanted to reassess the risk factors of life insurance policies and ensure that all relevant factors were identified. In addition, an AI was to evaluate the risk factors automatically and dynamically in the future. But it also had to calculate the insurance premiums dynamically (daily prices if necessary) and permanently adapt to the sales targets of the company as well as to legal regulations.

This was a challenge within the insurance industry, which HighPots had already met in the past. Technical and procedural solutions already existed that were implemented and made available by other insurance companies. Such experiences and existing products are sufficient to participate in a tender procedure.

At the end of the tendering period, only 3 of the 11 companies applying were left, including IBM and HighPots.
The remaining companies were invited to Japan. So we flew to Tokyo and drove on to Shinjuku.

Although IBM had its first appointment after ours, IBM Watson had somehow managed to establish a personal contact with Sompo beforehand. We had our software, a presentation and a rough project milestone plan with us. We didn’t leave any question unanswered.
And yet we had the feeling that the insurance company had already made a decision before our presentation even took place. As we learned afterwards, IBM had already introduced their Watson product during the tender phase. Sompo was obviously very impressed by the massive data analysis capabilities that IBM showed with Watson. So IBM finally got the contract and their project “Risk Calculation with IBM Watson” started at the end of 2014.

But meanwhile, the project has come to the conclusion that IBM Watson will not be used anymore. The insurance company does not want to make an official statement on that matter. But at a trade fair their project manager told us that IBM Watson was not able to bring any added value. It was obvious that the entire presentation had been obviously custommade by the IBM Marketing staff during the tender period in weeks of work specifically in regard to this tender. Otherwise, the project manager could not explain why IBM Watson had failed in the productive operation at the end. Millions of dollars had just been spent for simply nothing. Meanwhile HighPots is back on the boat; but unfortunately, the Japanese have used up almost their entire project budget, so now an extreme cost-cutting strategy has to be followed.

IBM Watson also fails at Munich RE and Swiss RE

It is not only in Japan that IBM Watson has failed in the insurance industry. IBM Watson has also failed in Germany/Munich at Munich Re, with the result that Munich Re has now removed IBM Watson. The IBM Watson project has also been shipwrecked in Switzerland at Swiss RE in Zurich. Apparently, IBM Watson is not delivering what it promises. For insiders of large corporations, IBM Watson is an example of outstanding marketing; since it is quite amazing how ordinary software for such a high price can still find customers over and over again. In addition to our own HighPots software products, there are of course other software products comparable to IBM Watson, such as Cogito, too.

Yet it is precisely the insurance companies with which IBM earns billions through the mainframe business; these insurance companies have been IBM customers of many years and it is precisely with these customers that IBM shouldn’t mess around under any circumstances. The business models and working methods of the insurance companies should actually be known to IBM. However, IBM Watson cannot keep its promises to insurers. It becomes a risky game if IBM does not change its marketing strategy.

IBM Watson also fails in the medical sector

Big Blue’s claims that Watson has a tremendously high added value in detecting disease is far more than an exaggeration. It almost falls into the category of  faking false pretences or simply “fake news”.
In September, the German business magazine “Wirtschaftswoche” reported about the news of the cancer research center MD Anderson that IBM Watson had failed and had been cancelled. The University of Houston (University of Texas) had loused up a total sum of US$60 million. This was another failure for IBM.

The intelligence of IBM Watson is doubted by authorities as well

A former IBM manager (Antonio Samaratini) now owns one of the best-known digital agencies in Italy. He is also advising the Italian Prime Minister. Three years ago, the ex-IBM manager began using IBM Watson in the Italian government. Meanwhile, Samaratini has seemingly become disillusioned. “IBM Watson cannot offer a realistic use of artificial intelligence in the governmental environment,” says Samaratini. This IBM Watson project also failed.

IBM Watson: Marketing, Marketing and again Marketing

Marketing raises needs. Needs in turn require trust. Trust is based in particular on three expectations:
– Expectation of competence – the expectation that the cooperation partner is competent in his domain.
– Expectation of integrity – the expectation that no covert or tactical strategies will be pursued.
– expectation of benevolence – the optimistic-open attitude towards other cooperation partners

IBM Watson Marketing takes advantage of their prospective customers’ expectations (and trust in the brand) to close a sale. But IBM is not able to meet the other two expectations. A failure of IBM Watson is therefore inevitable. Even if IBM should manage to improve its product at some point, the damage made to IBM Watson would still remain.

IBM Watson – less AI but high performance

According to our experts, IBM has already done its homework on performance. Data processing requests are processed at lightning speed. But artificial intelligence or machine learning are still missing at IBM Watson. For example, text analysis at IBM Watson works with dictionaries and lists of pre-defined terms. A text is “understood” by using statistical models only, but not semantic methods. However, only the combination of both components can finally be a success.

Problems with all AIs – it is not only IBM Watson that fails

If you look at all the common AIs, regardless of whether they are in B2C or B2B environments, they all do not really deserve the term “intelligent”. Apart from image recognition, nothing really works for them at all. Many companies are currently using Amazon’s Alexa to enable users to enter and output speech such as “”Liability insurance rates”” or “”Next flight from Frankfurt to London””. In order to make this possible, all terms, all phrases and all word combinations must be thought ahead. Some employees spend days working on such a task. Alexa cannot derive any word combinations of her own – the so called machine learning proves to be rubbish here! Amazon profits from many companies and consumers who teach Alexa the language with much effort (and it also charges money for it). In the background Amazon tries to vectorize words and sentences. But Google Now, Cortana and Siri are no better. People are obviously being manipulated to train the AIs. However, the little success achieved so far shows how much data must actually be required for training. Even the chatbots, which are often backed by the major providers of AIs, cannot have a reasonable conversation for 30 seconds without being exposed as a bot. The difference between the artificial neural networks behind AI’s and the biological neural networks is (still) immense. The human brain is still not fully explored. The digital imaging or digital copying of an organ that is still largely unexplored can only have a small probability of success. There is increasing evidence in current medical research that quantum mechanical effects are taking place in the human brain and which current IT cannot reproduce without the aid of quantum computers.

In any case, a bot that can search the FAQ database on the basis of a keyword entered by text or speech and can deliver the most likely result orally or in writing, can be built easily and inexpensively. You can either do this with commercial products from Nuance (Dragon Naturally Speaking Interfaces) or with the numerous open source text-to-speech and speech-to-text tools. With a cheap interface, e.g. to Babbel, the texts can also be translated into different languages in a reasonable quality.

IBM Watson also fails financially

Although IBM has made millions with Watson, the investment bank Jefferies states: “IBM’s revenues from Watson’s investments do not exceed investment costs, and that is not going to change in the medium term”.