Honest Fetch AI ICO review

Note: this is an English translation of the Russian article we wrote recently with an in-depth analysis of the Fetch AI ICO.

ICO Website | White Paper | ICO Chat | Medium | Github

The review was created by “Avocado Approves” community with the support of Golden borodutch Telegram channel. For the convenience of readers, the review is divided into six parts: product, advisors, team, partners, legal part and conclusion. We’d appreciate if the project team answered the questions in bold publicly.

FYI: this ICO will be held on Binance exchange. And that fact only is a weighty argument for many investors to buy tokens. Even more so if we keep in mind that the BTT recently got to that exchange too.

Worth noting that everything written below only represents the personal opinion of the author, which fully complies with the Fair Use laws and the First Amendment. It’s is in no way trading advice, and all the information was gathered from public sources.

Product

Fetch is a decentralized digital representation of the world in which autonomous software agents perform useful economic work.
— Product description in Fetch AI documentation

That’s one damn vague explanation that probably made your brain freeze. In short, it is a blockchain Protocol for automatic data exchange. To ease understanding of how that works, fetch provides a number of examples of using their product:

• Automated taxi ordering, flight bookings, and task management based on the user’s daily plans. To make that happen, the taxi service, the airline and the daily planner app must integrate the Fetch protocol. The user only needs to install the daily planner application.

Or does he? Does the user need to install your wallet and buy tokens on the exchange to start using the services of fetch? Does the same go for companies?

• Selling data about a vehicle activity and relocation to the geolocation service. That data would serve to inform other users about traffic jams, events and changes in the route. Data on when a car wipers are on or if there are many cars with windows closing at the same time at the place could be sold to a weather service. Weather and geolocation services then sell this data to the drivers traveling to the same location.
• Selling fitness bands data to researches for statistical studies in medicine.

That is, their Protocol and so-called “agents” are software intermediaries that connect sellers and buyers of data. You can compare Fetch with Google Adwords that collects data about users’ actions and sells the data to businesses. There is a difference though, which is that here users can selectively sell their data for tokens.

All this looks quite futuristic because the reality is not as smooth as the examples.

Technically, their network is a kind of hybrid blockchain (Smart Ledger), hybrid consensus (DAG PoS + uPoW) and sharding (Resource Lanes). The technical paper claims that the network can reach a million transactions per second. But the process of scaling, details on how the nodes and these technologies operate is described quite superficially.

How does your three-tier system relate to the nodes? And how will the nodes interact with each other?

Machine learning is said to optimize the data, control the learning agents, and predict the load of the network. But we found no technical details of these processes either.

How do you plan to make machine learning work in your system in a decentralized way? Will the neural networks be trained on the clients’ machines?

Fetch product consists of three parts:

1. Autonomous economic Agents, which are everything that can receive digital data: device sensors, equipment, databases, services, companies, and accounts. The Agents then automatically sell the data to those who need it.
2. Open Economic Base which is a framework for setting up your agent and interacting with others.
3. Smart register which is, in fact, a usual blockchain with sharding similar to Zilliqa. Although the Fetch team says that this is not a blockchain but a completely different technology, we did not see any differences.

Now we’ve got a couple of questions to the product.

• What is the legal side of the issue here? How do you plan to comply with the GDPR? We don’t need the team’s opinion; we need legal documents.
• What are the minimum technical requirements for agents and nodes?
• On what machines will the neural network be trained? What data will be its base?
• Who is responsible if an agent sends false data and somehow harms another user?
• How will you take into account all the variety of equipment, devices, and services for which you need to make the protocol, agents and learning algorithms?

Page 22 of FETCH: TECHNICAL INTRODUCTION, thanks to https://cryptosherlock.club/fetch_ai/ for pointing it out

• The second quarter of 2018 has long passed, and the promised technical details about how artificial intelligence will help to arrange the network’s security still nowhere to be seen. How come?

Previously, there were three use-cases on the Fetch website. Now you can only find those use-cases in the review from Binance. Two of them still raise questions:

Energy
Fetch is working on creating a fluid energy model to deliver the most effective energy solution to households without the friction of switching suppliers. Consumers could potentially change providers down to the minute or even by each appliance’s unique energy demands. Fetch is working with Warwick University on creating and deploying a live energy balancing simulation.

Supply Chain
Fetch is enabling the trillion dollar steel sector to autonomously and collaboratively self-manage. It is now able to optimize its supply chain from the raw materials to the finished product, giving it the opportunity to reduce costs and improve efficiency massively.
Fetch is also working on building collective prediction models to improve efficiency in cargo rail route in collaboration with a German team.

• Where can I find evidence of your work with Warwick University? And what are the current results?
• How exactly do you optimize your supply chain? And what technology stack do you use for that?
• On what principle “collective prediction” will work? Neural networks? If so, what type of neural networks will be used?
• Where do I find evidence of your collaboration with “the German team”? What is this team and what are the current results of your joint work?

According to the plans from the Fetch white paper, wallets, mobile applications, a public testnet should have already launched. Fetch has also failed to demonstrate how agents with artificial intelligence perform.

• Why do you fail deadlines?

Now the project team promises to launch a public testnet in the summer of 2019. The launch of the main network is planned for the end of the year. And as it stated on their website, all of their decentralized technologies will be open source.

Now let’s see what has been completed so far on GitHub:

• Smart register code published under the Apache license. It includes instructions on how to build and run the code on MacOS and Ubuntu.
• OEF framework code along with two SDKs for C++ and Python published.

We asked the project team why there was so little activity in the repositories on GitHub, to which the answer was:

We’re in full development mode and maintaining two repositories, the public one you’re seeing and the private one we currently push to.

This will be changing in the future once we release Fetch at its fullest. Until then, our public repo consists of very deliberate releases that are stable, and usable and show exactly what Fetch is and how some of Fetch will work. Our private repos are very busy. They’ll all be updated again soon.

In Binance’s review on Fetch, ten more private repositories were mentioned along with the number of contributors and commits.

It is also claimed that a private testnet of the project has been working for a long time, but there is no proof whether it is true.

Advisors

As usual, we wrote to each of the five project advisors. As responses are received, the review will be updated.

• Melvyn Weeks [in|fb] — Assistant Professor in Economics at the University of Cambridge, researching the application of Machine Learning to market pricing. Senior Economic Advisor to Ofgem, the UK’s Energy Regulator. Member of Smap Energy research laboratory, which develops SaaS solution that allows utilities to analyze energy consumption data using advanced machine learning techniques. We wrote to Melvin on LinkedIn and Facebook, still waiting for an answer.
• Steve Grand [in|tw] — An inventor of complex autonomous agents for nearly 40 years, creator of the Creatures artificial life games and proud father to a small robot now in the Science Museum, Steve has held research fellowships in artificial life, psychology, biomimetics, and creative technologies. He received a D.Univ from the OU and was made an OBE in 2000 for Services to Computing. He wrote two non-fiction books: «Creation: Life and how to make it» and «Growing Up with Lucy: How to Build an Android in Twenty Easy Steps». We wrote to Steve on Linkedin, still waiting.
• Dr. Niall Armes [in] — Dr. Armes is a world-leading biochemist, molecular biologist, and entrepreneur. He received his Ph.D. from the Imperial Cancer Research Fund in London for work on comparative genome structure. He subsequently founded TwistDx, serving as CSO and CEO prior to the company’s acquisition. We wrote to Niall on Linkedin, still waiting.
• Monique Gangloff [in] — Principal investigator/senior scientist at the University of Cambridge, Department of Biochemistry. Her research endeavors have resulted in more than 35 international peer-reviewed publications and one patent application. We wrote to Monique on Linkedin, still waiting.

Why does your project need an advisor on biotechnology?

• Jamie Burke [in] — Founder and CEO of Outlier Ventures. Jamie’s response on Linkedin was: “My company Outlier Ventures has been their lead investor and advisor for over 2 years. We are over 20 people. Please email matt@outlierventures.io”. We received no answers to the questions we had asked for.

Team

The team consists of 33 people. Like the advisors, everyone from the team lives in the UK. To assess the entire team, we are going to look into the experience of the founders and key engineers only.

• Humayun Sheikh [in|tg] — co-founder and CEO. In 2014, he founded the itzMe, a startup, which was focused on personalizing devices through machine learning; three years later the startup failed to launch and was shut down. In 2016, with funding from the Outlier Ventures, Sheikh founded uVue, a startup focused on automation of various tasks with the help of drones. Judging by the website traffic, financial statements and the fact that the fund removed them from its portfolio — the project did not become a success and mired in credit debts. By the way, now Outlier Ventures funds Fetch too. In LinkedIn, Humayun indicates that he was an investor and founder of DeepMind, which explores learning algorithms for Google’s artificial intelligence, tested in games. We have found no public records of Humayun connection to DeepMind. According to the UK Companies House, Humayun owns four operating companies in the following lines of business: waste management, a supply of metals, materials, and minerals. The website of one of the companies hasn’t been updated for a long time, and the others have not been working since the autumn of 2018 [1|2]. Financial reports indicate that these companies have a small income and massive debts starting from €4 500 000 up to €10 800 000. We could not find any company owned by Humayun that would produce functioning IT products.

Will the money from the FetchICO Fetch be used to cover loan debts Humayun’s companies?

Has Humayun ever launched a successful IT project, which eventually would benefit more than 10,000 people?

• Toby Simpson [in|tg] is a co-founder and CTO. He worked as a producer and Executive Director of a game series known as Creatures for nine years. After that Simpson began to develop his own games using AI: Diggers, Nicely Crafted and Global Effect. He worked as a CTO at Nice Tech for eleven years, where he developed a framework for online games-AliceServer, which later became the base for Time of Defiance game. In his, LinkedIn Simpson indicates that he worked as a head of the development Department at DeepMind for a couple of years, but we could not find any evidence. Also, Simpson worked as a CTO at Ososim, a company that provides IT training for organizations. In addition to Fetch AI, Simpson owns uVue. He also registered a small patent for “low power electric fence with detection mode.” Simpson’s Github has existed since 2011, has a PRO status, but is inactive and empty. His Beatbucket is empty, too. We couldn’t find any functioning projects or any code written by Mr. Simpson.

Where do I find evidence of Toby’s work with DeepMind as a head of development?

• Thomas Hain [in|tw|tg] is a co-founder and the Chief Science Officer. He was a researcher and lecturer at Cambridge University for six years. He has been teaching and studying speech and audio technology at the Department of computer science at the University of Sheffield for fifteen years now. He founded a startup for video analysis using AI-Koemei, which was bought by a banking service known as Crealogix. Hain is also the leader of the speech technology research group called WebASR. Hain participated in 183 research works on speech and audio technologies and has published more than a dozen research papers. He has a prominent scientific footprint on the Internet and an empty GitHub.
• Troels F. Rønnow [in] — head of Software Engineering. Rønnow received his Ph.D. for his work on carbon nanotubes. He tested a quantum computer and participated in the writing of 16 scientific papers on quantum physics. Rønnow worked as a researcher at Nokia for three years. He registered three technological patents for the company. Troels registered his GitHub account in 2011, but he hasn’t been very activity there. He mostly programs in C++ and Python. We couldn’t find and any senior positions in Troels’ work experience.

The project website says Troels is a co-author of 35 patents. Can you list them?

• Attila Bagoly [in|fb|tw] — software developer. Before working at Fetch, Attila was a graduate student in deep learning. And before that, he studied a lot, took a lot of courses on machine learning, and later he also taught these courses. He is active on GitHub, but mostly in private repositories. We could not find working projects by Bagoly, only his training materials.
• Khan Baykaner [in] — lead software engineer. According to LinkedIn, Baykaner has extensive training experience in psychoacoustics and a Ph.D.; he is a co-author of seven scientific articles and patents on the impact of sounds on hearing. He worked as a senior ML researcher at Nokia for almost years. We could find neither code no working projects by Khan Baykaner.
• Peter Bukva [in] — principal software engineer. Bukva worked on a mobile application for Bloomberg and the architecture behind Siemens. We could find neither code no working projects by Bukva.
• Joshua Croft [in|tg] — lead software engineer. On his linked page, Croft points out that he was a PlayStation designer and developer for almost two years. We could find neither code no working projects by Croft.
• Robert Dixon is a senior software engineer. He has a faint information footprint and an empty GitHub.
• Edward Fitzgerald [in] — lead software engineer. Fitzgerald indicates that he worked for eight years at Ericsson and a year at Nokia. He has a faint information footprint. We could find neither code no working projects by Fitzgerald.
• Nathan Hutton [in] — senior software engineer. He worked for two and a half years in the Wireless Technology Department at BAE Applied Intelligence, a cybersecurity services company. He is active on GitHub, but only in private repositories.
• Katie Lucas [in] — lead software engineer. Lucas’ LinkedIn page has an impressive list of work experience. She worked at Google as a security and privacy specialist for Google calendar for three years. Lucas also worked in Citrix, Hitachi, and Grapeshot dealing with security systems and data processing. She is inactive on GitHub inactive, but she ran tests for Fetch.
• Aristotelis Triadafyllidis [in] — software engineer. Previously, he worked as a game developer in a few small studios, but he was never employed for a long time. His Github’s empty.
• Pierre Wilmot [in] — software engineer. He claims to have a productive work experience at large studios in the position of a game developer. None of the projects he mentions works; his GitHub is empty.

All Fetch developers write in C++ and Python. But they are not on friendly terms with Open Source, which is why we could not check if any of them has any practical experience with machine learning or blockchain. It seems as if almost all developers’ accounts on GitHub were registered by one person specifically for the project.

Why did you publish empty GitHub accounts of your team on the website?

In addition to the three founders and eleven developers reviewed, the project has eight managers and eleven researchers in the following fields: machine learning, distributed systems, and cryptography.

Most of the team has a Ph.D. from Cambridge. Some of them published their scientific works. Only one of them though, Marco Favorito, has worked with machine learning or blockchain in practice.

Why are there so few real coders and so many University academicians in the project team?

Partners

The project has six confirmed partners:

• Outlier Ventures is a venture capital fund. For Fetch, they’re consultants and early investors.
• TokenMarket lists information about the Fetch ICO on their website. TokenMarket also consults and participates in multisig access to the collected funds.
• AiiN is a community of researchers studying the impact of artificial intelligence on business and society. Fetch helps them with their studies through Fetch blockchain and machine learning.
• ULedger blockchain for businesses. ULedger is supposed to share their IoT data with Fetch sometime in the future.
• MOBI — a community of researchers studying blockchain use in transportation. MOBI conducts joint research with Fetch.
• Blockchain for Europe is an Association for promoting blockchain projects in Europe.

Investment

According to Binance and Fetch’s token overview, they collected $974 and 975 24 596 ethers of private investments through SAFT in the spring of 2018. At that time, the exchange rate was about $16 500 000. It is unknown who exactly invested. We could not find an application to SEC about SAFT either.

• Have you reported to a financial regulator about SAFT investments? If you have, can we see your application on the regulator’s website?
• Who, how much and at what price bought tokens in the private round?

Binance website states that 21,125 ethers from those collected were converted into fiat currency during the 4th quarter of 2018 and 1st quarter of 2019. What it’s meaning is that before the 4th quarter of 2018 Fetch kept everything in Ether, which lead to the decrease of the initial sum in the total amount of ~$4 900 000.

• Why didn’t you convert the money to Fiat currency right away?

As of December 1, 2018, about 85% of the funds were spent on:

• 10% partnership programs;
• 10% marketing;
• 25% team;
• 45% development;
• 5% professional services;
• 5% other.
• What exactly was included in the cost of partnerships, professional services, and other things? Why are the cost of the team and the cost of development different points?

There are no exact figures on the costs, but we can approximately estimate them ranging from $4 000 000 $14 000 000.

• How have you been able to spend so much in nine months and have not released MVP?

Tokens of private investors are frozen for three months. ICO investors will receive their tokens 15 days after the token sale.

Token distribution amount: Fund-20%; founders-20%; token-sale-20%; future releases — 20%; mining — 10%; advisors — 10%.

• Why do founders and advisors need so many tokens?
• What do you mean by “future releases” and why do they need tokens?
• Why is there such a thing as a token fund?
• Where are the private investors?
• What motivation will miners have after they have run out of the designated tokens?

The team writes that the tokens of the advisors and founders will be unfrozen gradually, within three years. They can not prove this though since the token smart contract is private.

• Why don’t you publish your smart contract?

Fetch’s website published a rather superficial laudatory “audit” of their private smart contract. Judging by the search results, Hosho, the auditioning company, a high rating to every smart contract.

• Why did you choose Hosho as an auditor?

In total, the project will collect $6 000 000 on the token sale. Previously, the hard cap was listed as $30 000 000.

• Can we see a detailed estimate for each product? How many programmers, marketers, managers, scientists should be hired? In what fields? How much is Fetch going to pay each of the employees?

Legal part

Fetch.AI Foundation Pte Ltd, the company which is intended for ICO, registered in Singapore. Fetch.AI Limited, a product company, registered in the UK.

• Who are your lawyers?
• What will you do if there are legal claims for data collection from Google, Facebook or IBM?

The FET token is positioned as a Utility. The token will be needed to register agents and nodes in the network, exchange data, access various machine learning services, and exchange on ERC-20 for further sale.

• What do machine learning services that can be obtained with the FET token mean?

Countries where it is forbidden to participate in the Fetch token sale and other ICOs on Binance platform:

Afghanistan, Burundi, Belarus, Central Africa, Congo, China, Ethiopia, Guinea, Guinea-Bissau, Iraq, Iran, North Korea, Lebanon, Sri Lanka, Libya, Serbia, Sudan, Somalia, South Sudan, Syria, Thailand, Tunisia, Trinidad and Tobago, Ukraine, Uganda, USA, Venezuela, Yemen, Zimbabwe.

To participate, you must pass the verification on Binance and have a second level of identification.

• How do you plan to maintain the liquidity of your token?

Conclusion

Let’s take a brief look at the whole project.

• A data exchange protocol is a product that is difficult to understand, develop, and especially implement. What we see here is “blockchain and machine learning for everything,” which basically means for nothing.
• Among the advisors, there are cool specialists, but there isn’t anyone good at blockchain and high-load systems. Also, after almost a month, none of them have answered our questions.
• Fetch’s CEO backgrounds shoe nothing but millions in loan debts and two failed startups. That doesn’t look pretty, but these are the facts.
• The team has a great many academicians from Cambridge. But none of the software engineers have real projects in machine learning, blockchain or high load systems.

The most contradictory things seem to be:

1. The Outlier Ventures financed the project, despite having a bad experience with the Fetch’s CEO ’s previous project, uVue. uVue, being focused on drones, spent all the invested money and got drowned into credit debts. It is evident that the Fund lost money.
2. Binance didn’t pay attention to the fats written in their own report. The founders of Fetch are bad at managing investors’ money, given the massive spending in the short term and the lack of results. And at the same time, Binance allowed this ICO to its platform.

This article partly used Fetch reviews from Binance, Sherlocks, ICODrops, and Zavodil.

Update February 19, 2019:

• The administration of the project chat decided to ban anyone who somehow praises the review or asks questions from it.
• The Technical Director wrote a chat response to the conclusion of this review, where he mentioned the credibility of his team and argued that the review was not deep enough. He did not give evidence of his statements and did not answer the questions.

Update February 20, 2019:

The official response of the technical director in correspondence with Nikita Kolmogorov