Neural network of 1000 Indians

Technology startups often overestimate their capabilities related to artificial intelligence.

Lina Medvedeva, XX2 century

Startup Engineer.ai He stated that he uses artificial intelligence technology to automate the development of mobile applications, but several former and current employees of the startup report that the company exaggerates its capabilities to attract customers and investors.

Since artificial intelligence (AI) technology is quite complex and vaguely defined, it can be difficult even for experts to determine whether it is actually implemented. Nevertheless, money flows into this sector and many startups can say that they use artificial intelligence to attract investments or corporate clients, even if such statements are difficult to verify.

Venture capital firms have almost doubled funding for AI startups since 2017 to $31 billion. According to PitchBook, the number of funded startups with .ai in domain names has more than doubled in recent years.

In July, the Japanese technology conglomerate SoftBank Group Corp. announced the opening of an investment fund focused on artificial intelligence, its expected investment fund is $ 108 billion.

Last year, the company Engineer.ai With offices in London and Los Angeles, has raised $29.5 million from investors, including from Deepcore Inc., a subsidiary of SoftBank. Other investors include the Zurich venture capital company Lakestar, the first investor Facebook Inc. and Airbnb Inc., and Singapore-based Jungle Ventures.

Engineer.ai claims that its "artificial intelligence with human participation" allows anyone to create a mobile application by clicking on the menu on the company's website. Users can choose existing apps similar to the one they need, such as Uber or Facebook. Then Engineer.ai creates an application – pretty much automatically, as the company claims. It seems to make the process cheaper and faster than conventional application development.

"We have created software and artificial intelligence named Natasha, which allows anyone to create custom software, for example, to order pizza," said the founder Engineer.ai Sachin Dev Duggal in an interview in India last year. Since most of the code underlying popular applications is of the same type, "artificial intelligence with human participation" can help create new applications automatically. Approximately 82% of the company's newly developed applications were "created offline within one hour" using technology developed by Engineer.ai , Duggal reported.

However, consideration of the facts, which were told by current and former employees Engineer.ai , indicates that the company does not use artificial intelligence to write application code, contrary to what it claims. Employees said that the company uses the work of human engineers from India and other countries to do most of this work. Her claims about the artificial intelligence created are overstated even in the light of the "Fake while you finish" mentality common among technology startups.

Engineer.ai only in the last three months has he started creating the technologies necessary to automate the writing of applications, according to a person familiar with the company's activities. He added that it will take the company more than a year to start using any artificial intelligence for basic maintenance.

Press Secretary Engineer.ai and Sachin Dev Duggal, who calls himself the "chief magician" of the company, told reporters The Wall Street Journal that they "clearly understand what they are doing," emphasizing that the company uses technologies characterized as artificial intelligence with human participation.

A SoftBank spokeswoman declined to comment.

In the lawsuit, which was not previously reported, the former chief commercial officer Engineer.ai Robert Holdheim questioned the technical skill of the company. According to his complaint of unlawful dismissal, filed in February in The Los Angeles Superior Court, Sachin Dev Duggal, told Holdheim: "Any technology startup lies for the sake of obtaining financing – this is money that allows developing this technology." Holdheim added that Duggal "was telling investors that Engineer.ai 80% completed product development when in fact he just started developing it."

Robert Holdheim declined to comment on the lawsuit because, in particular, he was fired after he got into a confrontation with Duggal over the misuse of investor funds. Engineer.ai he disputed these allegations in a subsequent statement, and a representative of the company told reporters that he could not comment on this issue during the trial.

When asked how the company uses artificial intelligence, its representative replied that Engineer.ai calculates deadlines and prices for customers absolutely automatically. Part of this process is natural language recognition. The company also uses a resolution tree to distribute tasks among developers.

Several current and former employees stated that some calculations of prices and deadlines are made by conventional software, not artificial intelligence, and most of the work is done manually by employees. These people said that the company lacks natural language processing technology and that the resolving tree used in the company cannot be considered as artificial intelligence.

Referring to the resolving tree is a "stretch," says Luka Crnkovic-Friis, CEO of the Swedish company Peltarion, which sells software for creating deep learning artificial intelligence systems. "If you tell customers you're using artificial intelligence, they probably won't expect it to be 1950s technology. Permissive trees are an old and simple technology."

Engineer.ai draws attention to the statement published on the company's website, according to which "on average about 60%" of the software is produced by machines, and the rest by people. The statement does not explain exactly how this share of production was produced by machine. The representative said that these details are the property of the company, and declined a request for clarification.

Engineer.ai there is a lack of a group of specialist engineers with significant experience working with machine learning or artificial intelligence, the company could only point to one such employee on its staff.

In a subsequent statement on the website Engineer.ai It was said that artificial intelligence experts were difficult to find and hire, and that some recently hired specialists studied machine learning and artificial intelligence. Separately, the company described in detail the experience of three team members in the field of data science and other disciplines, but did not name them by name.

Representative Deepcore stated that it is fully confident in the vision of Duggal and his team. A representative of Jungle Ventures said he was proud of his investments in Engineer.ai , adding that "the landscape of artificial intelligence is a wide spectrum." A Lakestar spokeswoman said she was also confident in Engineer.ai and her team, because "the growth of artificial intelligence does not happen overnight." She said that Engineer.ai she showed great caution by presenting her technologies to the company Lakestar and other investors.

In Europe, startups with a mention of artificial intelligence in the description of what they do have attracted 15-50% more funding than startups with other software, according to an analysis of 2,830 technology startups conducted by the MMC Ventures investment fund. This analysis states that about 40% of companies classified as AI startups have not provided any evidence of the use of artificial intelligence in their products.

"I think this percentage is even higher," said Vasile Foca, managing partner of Talis Capital, a London–based venture capital firm that supports startups with artificial intelligence. "You will get three to four times more interest from venture investors if you claim that you have artificial intelligence or your decision depends on artificial intelligence."

Crnkovic-Friis, not related to Engineer.ai , says that many startups usually find that creating artificial intelligence is harder than expected. Among other things, it may take years to collect and mark up the data needed to train the machine learning algorithms underlying such technologies.

According to Crnkovich-Friis, to teach new algorithms to an application developer like Engineer.ai , it is necessary to collect thousands of requests from customers and combine them with the code created by engineers. Several people familiar with the work Engineer.ai they say that the necessary data has not yet been collected.

Representative Engineer.ai He said that the company has collected more than 600 million records to help create artificial intelligence based on customer interaction.

According to Crnkovich-Friis, some companies use cheap human labor as a temporary way out before implementing real machine learning algorithms.

Portal "Eternal youth" http://vechnayamolodost.ru