Thread by @MTorygreen

AI may soon destroy the world…

…but not in the way you might think

Our only chance to stop it may be to use #crypto

Projects like @oceanprotocol, @Singularity_NET, @DeepBrainChain, @CTXCBlockchain, @numerai and @Fetch_ai are leading the charge

Here’s why

🧵

👇

2/

This thread will cover the following:

• What is AI?

• How does AI work?

• The problem with AI

• How blockchains can fix AI

• The decentralized AI ecosystem

• The long-term potential of AI

3/

🔶 What is AI?

AI is a generic term that refers to computer programs that are advanced enough to perform tasks that typically require human intelligence to complete

In practice, when people speak about “AI” today, they’re talking about Machine or Deep Learning

4/

Although Machine and Deep Learning may seem like magic, they’re not – they’re just advanced pattern recognition tools

They analyze vast amounts of information, find hidden patterns and use the analysis to solve problems and make complex decisions

5/

That doesn’t make them any less impressive though, as computers can use “pattern recognition” to:

• Create art

• Write novels

• Build apps

• Converse with humans

• Drive cars

And a whole lot more!

6/

🔶 How does AI work?

Most deep learning is done using artificial neural networks (“ANNs”)

ANNs are structured to mimic the function of the human brain - they are composed of neurons, which are arranged in layers and learn through a process known as backpropagation.

7/

🔹 Neurons

Neurons are the brain’s information messengers. They receive information through dendrites, and determine if this information is important

If it is they “activate” and pass along the information to other neurons via a structure known as an axon

8/

Artificial neurons behave in a similar manner. They receive information, decide whether it’s important, and pass signals along to other artificial neurons.

9/

🔹 Layers

The neurons in an ANN are arranged into three layers:

• Input Layer: Receives information from the outside world

• Hidden Layers: Performs calculations

• Output Layer: Returns the outcome

ANNs can have dozens to hundreds of hidden layers

10/

These layers work through specialization

Let’s say you wanted to detect a face:

• One layer might detect edges and basic shapes

• The next would see if these shapes were eyes, ears or noses

• The third layer would see if these eyes, ears or noses made a face

11/

🔹 Backpropagation

Most artificial networks learn in a manner that is similar to their biological counterparts – through feedback

12/

For instance, imagine a child learning to play basketball

If his first shot was short, he would throw the ball harder next time

If the second missed to the right, he would then aim a little left

He would continue this through trial and error until he made a few baskets

13/

ANNs work the same way, through a process called “backpropagation”:

• The ANN makes a prediction and compares the results to the target output

• It calculates how far off it was, adjusts and tries again

• It continues this process until it gets it right

14/

🔶 Problem

In its current form, AI has little risk of becoming “sentient” and taking over the world

But it may pose an even bigger short-term threat – it may enable “Big Tech” to become exceedingly powerful and take over the world

15/

To understand why, it’s important to realize that AI needs three “raw materials” to function:

• Data

• Models

• Compute

All 3 of these resources are dominated by Big Tech

16/

🔹 Data

Deep learning models require massive amounts of data to train.

For example, OpenAI’s GPT-3 model required 45TB of text data (equivalent to 15 billion printed pages)

17/

Guess who controls the majority of data in the world today?

That’s right – “Big Tech” companies like Google, Amazon, Microsoft and Facebook

18/

🔹 Models

Significant amounts of human capital are needed to create and maintain AI

A typical ML or DL “flow”, involves several steps that need human oversight:

1. Data Collection

2. Data Cleaning

3. Model Selection

4. Model Training

5. Model Optimization

19/

Unfortunately, this talent is not spread around equally

The top data scientists are often either employed by Big Tech – as they are the only ones with the resources to support the armies of PhDs needed to build, train and optimize machine and deep learning models

20/

🔹 Compute

AI is getting more expensive.

Recent advances such as OpenAI’s GPT-3 required 3,640 petaflop days to train – over 500,000x more than similar breakthroughs a decade ago

21/

This trend shows few signs of stopping, and many researchers believe that the demand for computing resources will continue to grow at 700% per year

(which is much faster than the supply of resources, which only doubles every two years)

22/

Right now, Big Tech companies are the only ones with the resources to build the computers required for AI

For instance, Facebook is building the AI Research SuperCluster (RSC), a supercomputer that is expected to process up to 5 exaflops (>3x the fastest computer today)

23/

Big Tech has a monopoly on all three of these resources

And this creates a vicious cycle

More resources leads to better AI, better AI increases profits and more profits allow for the acquisition of more resources

24/

Indeed, the top four technology companies already control over 70% of the internet, and AI has the potential to make this problem much worse

25/

🔶 Solution

The way to stop Big Tech is to break their monopoly on the “raw materials” that create AI

In particular, we need to build decentralized:

• Data marketplaces

• Model marketplaces

• Supercomputers

• AI-enabled L1s

• AI-enabled dApps

26/

🔶 Decentralized AI ecosystem

Six protocols leading way in decentralized AI are:

• Ocean Protocol (@oceanprotocol)

• Singularity Net (@Singularity_NET)

• Numerai (@numerai)

• Deep Brain Chain (@DeepBrainChain)

• Cortex (@CTXCBlockchain)

• Fetch AI (@Fetch_ai)

27/

🔹 Ocean Protocol

Founded by @BrucePon in 2017, Ocean is a decentralized marketplace for data

Data owners can sell access to a variety of information types including public data, private data and synthetic data

Consumers can purchase or license this data

28/

Decentralized data markets such as Ocean bring a variety of benefits

• Privacy (consumers control their online identity and information)

• Profits (Users can monetize data)

• Lower Transaction Costs

• Ubiquitous Access

• Security (No single point of failure)

29/

Transactions are facilitated through either data NFTs or datatokens – NFTs allow a user to purchase full ownership rights a dataset while tokens only grant access to the data

Like a DeFi DeX, pricing is determined through an AMM protocol

30/

#OceanProtocol currently has a $99M market cap and $227M FDV

It leverages the $OCEAN token which currently trades at $0.16

31/

🔹 SingularityNET

Founded by @bengoertzel in 2017, SingularityNET is a decentralized marketplace for models

Data scientists can “rent” out ML and DL models to individuals and companies that need them

Customers pay in $AGIX and can leave reviews on the models

32/

Decentralized marketplaces provide many advantages to all market participants, including security, trust, privacy, lower transaction costs, and transaction integrity

Most importantly, they give data scientists an option to make money without working for Big Tech

33/

One unique feature of SingularityNET is that the models are interoperable

For instance, you could combine a model that translates Chinese to English to one that performs text to speech conversion in English to get a tool that translates written Chinese into spoken English

34/

#SingularityNET currently has a market cap of $59M and FDV of $104M

It leverages the $AGIX token, which currently trades at $0.05

35/

🔹 Numerai

Founded by @richardcraib in 2015, Numerai is an AI-based hedge fund that crowd-sources decisions through a weekly stock picking “tournament”

36/

Each week, the platform releases an encrypted set of market data

Data scientists use this data to create predictive ML and DL models

Numerai chooses and combines the best models, uses them to make investment decisions and rewards the creators with payouts in Bitcoin

37/

While this is an interesting approach, it’s not exactly novel, as Google’s Kaggle has been hosting predictive data science tournaments for years

What is novel is how Numerai approaches a problem known as overfitting

38/

Overfitting is a common problem in machine learning

It occurs when models match historical data too well, finding patterns that aren’t really there

While overfit models look like they are accurate, they tend to perform poorly when given new data.

39/

Numerai reduces the tendency to overfit by required data scientists to stake the protocol’s native token $NMR as collateral

If the models perform well, the creators receive payouts in dollars or BTC

If they perform poorly, the collateral is burned

40/

#Numerai currently has an $84M market cap and $157M FDV

It uses the $NMR token which currently trades at $14.32

41/

🔹 Deep Brain Chain

Founded in 2017 by a trio of AI veterans, DeepBrain Chain is a shared computing platform that allows users to buy and sell idle computing resources

In particular, the platform focuses on graphical processing units (GPUs), which are a primary tool of AI

42/

Shared computing offers several benefits:

• More Processing Power: DBC leverages a global network

• Faster Speeds: up to 20% faster than centralized networks

• Lower Costs: Up to 70% cheaper than competitors

• More Security: Multiple nodes provide redundancy

43/

#DeepBrainChain currently has a market cap of $5M and FDV of $16M

It uses the $DBC token, which currently trades at $0.0016

44/

🔹 Cortex

Traditional L1s such as Ethereum don’t have the bandwidth to run complex machine and deep learning calculations on-chain

This puts them at a significant disadvantage to Web 2.0 competitors

Cortex is an AI-enabled L1 that seeks to overcome that problem

45/

Founded by @zchenzchen, the network seeks overcome the limitations of traditional smart contract platforms by performing complex AI calculations off-chain, and then bringing only the results on-chain

To satisfy the computing requirements of AI, all Cortex nodes run on GPUs

46/

The Cortex ecosystem is comprised of a network of smart contract developers, AI engineers and miners, and contains:

• A library of AI models

• AI-ready smart contracts

• System for independent verification

47/

#Cortex currently has a $35M market cap and $50M FDV

It uses the $CTXC token, which currently trades at $0.17

48/
🔹 Fetch AI

Founded in 2017 by @HMsheikh4, FetchAI is building a decentralized network of autonomous “agents” that can perform real-world tasks

These agents are intelligent programs that act on their owner's behalf and make their own decisions

49/

Some early use cases for Fetch’s agents include:

• Optimizing trading for financial services users

• Reconfiguring public transport networks

• Supporting smart cities’ ability to adapt to citizen behavior

• Disintermediate the gig economy to remove middlemen

50/

#FetchAI currently has a $75M market cap and $116M FDV

It uses the $FET token, which currently trades at $0.10

51/

🔶 Potential

The potential for AI is immense

Many experts predict that it will increase productivity and reduce costs by orders of magnitude

As such, some researchers – such as Ark’s Cathie Wood – estimate that the market for AI will grow to over $80T by 2030

52/

But right now, all signs point to Big Tech firms – such as Google, Facebook, Amazon, Apple, Microsoft and Alibaba – owning this technology

That why the development of decentralized AI is so important!

53/

Important Disclaimer: None of the aforementioned projects are “calls” or “picks”

While I’m bullish on their “stories”, I haven’t done in-depth research into their management teams, performance, tokenomics, etc…

So if you’re going to invest, PLEASE do your own research

I hope you've found this thread helpful.

Follow me @MTorygreen for more fundamental analysis on Web3 protocols.

Like/Retweet the first tweet below if you can: