“AI” is becoming a construct that has been the subject of increasing attention in technology, media, business, industry, government and civil life during recent years.

Today’s AI is the subject of controversy. You might have heard about narrow/weak, general/strong/human level and super artificial intelligence, or about machine learning, deep learning, reinforced learning, supervised and unsupervised learning, neural networks, Bayesian networks, NLP, and a whole lot of other confusing terms, all dubbed as AI techniques.

Many of the rules and logic-based systems that were previously considered Artificial Intelligence are no longer AI. In contrast, systems that analyze and find patterns in data are dubbed as machine learning, widely promoted as the dominant form of AI.

What is Wrong with Today’s AI, Its Chips and Platforms?

All the confusion comes from an anthropomorphic Artificial Intelligence, AAI, the simulation of the human brain using artificial neural networks, as if they substitute for the biological neural networks in our brains. A neural network is made up of a bunch of neural nodes (functional units) which work together, and can be called upon to execute a model.

Thus, the main purpose in 2021 is to provide a conceptual framework to define Machine Intelligence and Learning. And the first step to create MI is to understand its nature or concept against main research questions (why, what, who, when, where, how).

So, describe AI to people as an AAI or augmented intelligence or advanced statistics, not artificial intelligence or machine intelligence.

Now, re the levels of AAI applications, tools, and platforms?

Lets focus only on “AAI chips”, forming the brain of an AAI System, replacing CPUs and GPUs, and where most progress has to be achieved.

While typically GPUs are better than CPUs when it comes to AI processing, they usually fail, being specialized in computer graphics and image processing, not neural networks.

The AAI industry needs specialised processors to enable efficient processing of AAI applications, modelling and inference. As a result, chip designers are now working to create specialized processing units.

These come under many names, such as NPU, TPU, DPU, SPU etc., but a catchall term can be the AAI processing unit (AAI PU), forming the brain of an AAI System on a chip (SoC).

It is also added with 1. the neural processing unit or the matrix multiplication engine where the core operations of an AAI SoC are carried out; 2. Controller processors, based on RISC-V, ARM, or custom-logic instruction set architectures (ISA) to control and communicate with all the other blocks and the external processor; 3. SRAM; 4. I/O; 5. the interconnect fabric between the processors (AAI PU, controllers) and all the other modules on the SoC.

The AAI PU was created to execute ML algorithms, typically by operating on predictive models such as artificial neural networks. They are usually classified as either training or inference generally performed independently.

AAI PUs are generally required for the following:

• Accelerate the computation of ML tasks by several folds (nearly 10K times) as compared to GPUs
• Consume low power and improve resource utilization for ML tasks as compared to GPUs and CPUs

Unlike CPUs and GPUs, the design of single-action AAI SoC is far from mature.

Specialized AI chips deal with specialized ANNs, and are designed to do two things with them: task-designed training and inference, only for facial recognition, gesture recognition, natural language processing, image searching, spam filtering, etc.

In all, there are {Cloud, Edge, Inference, Training} chips for AAI models of specific tasks. Examples of Cloud + Training chips include NVIDIA’s DGX-2 system, which totals 2 petaFLOPS of processing power, made up of 16 NVIDIA V100 Tensor Core GPUs, or Intel Habana’s Gaudi chip or Facebook photos or Google translate.

Sample chips here include Qualcomm’s Cloud AI 100, which are large chips used for AAI in massive cloud datacentres. Another example is Alibaba’s Huanguang 800, or Graphcore’s Colossus MK2 GC200 IPU.

Now (Cloud + Inference) chips were used to train Facebook’s photos or Google Translate, to process the data you input using the models these companies created. Other examples include AAI chatbots or most AAI-powered services run by large technology companies. Here is also Qualcomm’s Cloud AI 100, which are large chips used for AAI in massive cloud datacentres, Alibaba’s Huanguang 800, or Graphcore’s Colossus MK2 GC200 IPU.

(Edge + Inference) on-device chips examples include Kneron’s own chips, including the KL520 and recently launched KL720 chip, which are lower-power, cost-efficient chips designed for on-device use; Intel Movidius and Google’s Coral TPU.

All of these different types of chips, training or inference, and their different implementations, models, and use cases are expected to develop the AAI of Things (AAIoT) future.

How to Make a True Artificial Intelligence Platform

In order to create a platform neutral software operating with world’s data/information/content which could run/display properly on any type of computer, cell phone, device or technology platform, the following are required:

• Operating Systems.
• Computing/Hardware/Cloud Platforms.
• Database Platforms.
• Storage Platforms.
• Application Platforms.
• Mobile Platforms.
• Web Platforms.
• Content Management Systems.

The AI programming language should act as both the general programming language and computing platform. Its applications could be launched on any operating system and hardware, from mobile-based operating systems, as Linux or Android, to hardware-based platforms, from game consoles to supercomputers or quantum machines.

The post HOW TO CREATE AN ARTIFICIAL INTELLIGENCE GENERAL TECHNOLOGY PLATFORM appeared first on Artificial Intelligence.

In their machine learning-based capsid diversification strategy, the team focused on a 28 amino acid peptide within a segment of the AAV2 VP3 capsid protein that exposes the AAV capsid to neutralizing antibodies produced by individuals and thus can be the cause of an immune response against the virus. More purple colored portions of this peptide are buried deeper in the capsid, while yellow parts are exposed on the virus’ surface. Credit: Wyss Institute at Harvard University (original by Drew Bryant).

Adeno-associated viruses (AAVs) have become promising vehicles for delivering gene therapies to defective tissues in the human body because they are non-pathogenic and can transfer therapeutic DNA into target cells. However, while the first gene therapy products approved by the Federal Drug Administration (FDA) use AAV vectors and others are likely to follow, AAV vectors still have not reached their full potential to meet gene therapeutic challenges.

First, currently used AAV capsids – the spherical protein structures enveloping the virus’ single-stranded DNA genome which can be modified to encode therapeutic genes – are limited in their ability to specifically hone in on the tissue affected by a disease and their wider distribution throughout the human body causes them to be diluted. And secondly, patients’ immune systems, after having been exposed to a similar AAV virus, can produce neutralizing antibodies that, even at low levels, can destroy AAVs upon re exposure (neutralization), blocking the delivery of their therapeutic DNA payloads.

To overcome this neutralization problem, researchers are engineering enhanced AAV capsids they hope to be able to evade the immune system. Currently used methods, including “directed evolution” strategies that fast-track the evolution of a protein in laboratory conditions, only can create a limited diversity of capsids with most of them still resembling the naturally occurring AAV variants known as serotypes. However, it remains difficult to generate sufficient diversity using this approach without losing other desired functions of the capsid, such as their stability or ability to bind to specific cell types.

Now, a new study initiated by Wyss Core Faculty member George Church’s Synthetic Biology team at Harvard’s Wyss Institute for Biologically Inspired Engineering, and driven by a collaboration with Google Research has applied a computational deep learning approach to design highly diverse capsid variants from the AAV2 serotype across DNA sequences encoding a key protein segment that plays a role in immune-recognition as well as infection of target tissues. AAV2 is the most-studied serotype and has been used in the first FDA approved gene therapy, to treat a blinding disease.

Starting from a relatively small collection of capsid data, the team trained multiple machine learning methods and used them to design 200,000 virus variants. 110,689 of these variants produced viable AAV viruses. Between any two naturally occurring AAV serotypes, 12 amino acids within this segment are expected to differ. The team’s effort produced more than 57,000 variants that exhibited much higher diversity than this, some containing up to 29 combined substituted or additionally inserted amino acids. The findings are published in Nature Biotechnology.

“Our approach achieves the highest functional diversity of any capsid library thus far. It unlocks vast areas of functional but previously unreachable sequence space, with many potential applications for generating improved viral vectors, like AAVs with much reduced immunogenicity and much improved target tissue selectivity, and also for highly efficient gene therapies,” said last-author Eric Kelsic, Ph.D., who started the project with Church, Ph.D., and co-founded the startup Dyno Therapeutics where he is now CEO. Dyno Therapeutics’ mission is to develop advanced gene therapy delivery vehicles by employing cutting-edge artificial intelligence (AI) approaches.

Using multiple design strategies, the team first generated smaller data sets on which they could train several machine learning models. These were collections of AAV capsids with variable numbers of mutations introduced in a 28 amino acid segment of the AAV2 VP3 protein that forms part of the capsid and exposes it to neutralizing antibodies. A high-throughput method enabling the synthesis of mutated capsid sequences and in vitro experiments for testing which ones efficiency produced viable stable capsids, provided a highly effective test bed for their overall approach. The results from this first experimental study then were used by the team as training data for three alternative machine learning models that generated much larger numbers of diverse capsid variants to be tested with a final validation experiment.

A central bottleneck in the creation of diverse AAV capsids and variants that can evade neutralization is the production of capsids that remain stable: most of the variants will fail to assemble into functional capsids or package their AAV genomes. “The deep neural network models that we deployed with our Google collaborators accurately predicted capsid viability across extremely diverse variants. Reaching this level of diversity in the capsid segment is an important milestone that we can build on to find immune-evading capsids for gene therapy,” said co-first author Sam Sinai, Ph.D., a former graduate student of Church who joined Kelsic’s team at the Wyss Institute and is a co-founder leading the machine learning team at Dyno Therapeutics. “And we can take similar approaches to create AAV capsids with much improved tissue selectivity.”

In 2019, a former Wyss team including Kelsic, Sinai, and their mentor Church published a related approach in Science in which they mutated one by one each of the 735 amino acids within the entire AAV2 capsid in different ways. What they called a “wide” search resulted in a large AAV library that identified changes affecting AAV2’s viability and its “homing” potential to specific organs in mice, as well as a previously unknown accessory protein that binds to cell membranes and which was hidden within the capsid-encoding DNA sequence. In their previous study, the researchers used a simple experimental model to optimize the tissue targeting ability of the virus.

“This new study involving machine learning models developed with Google Research nicely complements our earlier work in that it focuses on a small, but very important, region of the AAV capsid with an unprecedented resolution,” said co-corresponding author Church. “It shows that neural networks combined with the high-throughput synthetic testing developed in our lab is changing the way we design gene delivery vehicles and protein drugs.” Church is the lead of the Wyss Institute’s Synthetic Biology platform where the project was started, and Professor of Genetics at Harvard Medical School and of Health Sciences and Technology at Harvard and MIT.

“This work gives a glimpse into the future as artificial intelligence approaches, such as machine learning, are opening up vast new design spaces that enable the development of entirely new drugs and drug delivery approaches for combating innumerable challenges to human health. It also highlights the Wyss Institute’s commitment to computational problem-solving in areas where new therapies are desperately needed,” said Wyss Founding Director Donald Ingber, M.D., Ph.D., who is also the Judah Folkman Professor of Vascular Biology at Harvard Medical School and Boston Children’s Hospital, and Professor of Bioengineering at SEAS

Reference: Bryant DH, Bashir A, Sinai S, et al. Deep diversification of an AAV capsid protein by machine learning. Nat Biotechnol. 2021. doi:10.1038/s41587-020-00793-4.

This article has been republished from the following materials. Note: material may have been edited for length and content. For further information, please contact the cited source.

The post Using Machine Learning To Create Better Gene Therapies appeared first on Artificial Intelligence.

NEW YORK—Google’s DeepDream has generated artwork; the What-If Machine created the characters and story for a West End Musical; music composed by programs was performed in the London Symphony in 2012. We talk about jobs that may be lost to automation, but there is scant attention paid to who owns the intellectual property created by machines.

Artificial intelligence technology no longer allows us the luxury to vacillate. With the advent of AI, software and computers will be creating a number of programs and original works. 

But in many parts of the world, IP laws have not kept pace with the technology. US copyright and patent statutes for instance traditionally required humans as authors or inventors in order for works to be protected under copyright or patent law.

Part of the problem may be historical—copyright laws for instance, were crafted to address printing press technology (that enabled human beings to copy at little to no cost) and not artificial intelligence and the ability for software programs to generate more programs.

The latter point has a real world impact—if the IP for programs generated from code belongs to the code, it could impact the Indian IT industry, and cost millions their jobs.

Other countries such as the UK and the European Union on the other hand have modified their laws, or are considering proposals to do so in order to address the advances, and breakthroughs in robotics technology. The UK for instance, has done away with the requirement of the human author and is conceivably open to awarding copyright protection to bots created by bots.

The UK, when introducing software programs into the Copyright, Designs and Patents Act in 1988 specifically did away with the requirement of “human author” (when recognizing which works would be eligible for copyright). 

In 2017, the Committee on Legal Affairs of the European Parliament released a study on AI and asked for the elaboration of criteria for “own intellectual” creation for copyrightable works produced by computers or robots; thus, paving the way for copyright protection for AI. 

Who owns the AI?

The problem of who will be the “owner” of the programs created by programs poses a thorny issue. Some folks point to programs such as Google’s DeepDream art, and argue that the program was a “tool” or paintbrush that enabled human authors’ vision to be manifested and so, the human author would own the IP. Others argue that the programmer who wrote the software or algorithm should be the author and copyright owner.

The law thus far has been silent on who owns the IP when the machine output cannot be predicted by the humans involved. There is a third group which includes technology savants like Elon Musk, who argue that AI created work should be in the public domain, owned by no one. 

Will contracts supersede the law?

While we wait for the laws in connection with IP ownership of AI to be developed or crafted, people may (by contract) have unwittingly given up that right. Right now, most employment contracts include a “work for hire” concept. As a result, most employers that use software programmers, writers etc. as employees or contractors include a wide provision that states all work produced during the term of employment or contract will be owned by the employer. The salary or fees paid to create the work is considered adequate compensation, including to transfer ownership in IP rights. 

On the face of it, the above may not seem worrisome or particularly important. However, it becomes significant because of the additional complication presented by the following contracting practice. 

Why should India’s technology industry worry

As a corollary to the above ‘work for hire’ employment contracts, large corporate customers particularly US clients have consistently insisted that all Indian technology corporations providing services to the US corporation transfer all the IP produced during the provision of, and in connection with the services. Indian technology corporations by and large have not objected to this.

Indeed for decades Indian technology corporations have transferred all IP to the client, believing the secret-sauce was the know-how or knowledge of how to implement the software programs. Hence, by inserting locks or restricting US clients from hiring their software programmers, Indian technology industry was content. 

This is about to change. Once copyright protection is recognized for AI code, no one else will be allowed to copy that code for a long time (the life of the author + 50 or 70 years depending upon the jurisdiction). Similarly, if a patent is granted for a code, no one else will be allowed to use that code for a long time (20+ years from the date of patent application).

Hence, if AI is granted formal IP protection, and the US clients own all the formal IP, they could take advantage of the protection offered by copyright and patent laws and be able to prevent Indian companies from performing the same services for other clients or corporations. In other words, if the Indian companies used the same code for other clients they could be sued for copyright or patent infringement. Especially so, because the U.S. has been comfortable recognizing patents for algorithms in connection with AI. 

Steps now needed

To continue to ride the technology wave in the AI era, Indian industry will have to adapt quickly and dramatically modify its contracting and negotiating practice. Companies must negotiate hard to retain the formal IP in order to be able to continue to operate their service lines in future. Hence, the risk presented by AI will not only be the loss of US and EU jobs as a result of computerization – estimated by Oxford University’s study as 47% and 54% of the US and EU workers’ jobs and the trickle-down effect on India but also from being able to write programs or perform work for other corporations as a result of AI IP they have written and handed over to their clients. 

It is an existential moment for the industry. At risk is not only half the 3.9 million Indians’ jobs  or 70% of the Indian IT workforce (as a result of automation) but the $155 billion industry that has been the engine for India’s economy and global image. It is not all dark – if Indian firms were to change strategy and to negotiate to own the IP they create, India may well succeed in riding the technology 2.0 or AI wave and epitomize the latter part of Hawking’s prediction:

“Whereas the short-term impact of AI depends on who controls it, the long-term impact depends on whether it can be controlled at all.” – Stephen Hawking

Aarthi Anand is a leading technology attorney, Vice President at J.P. Morgan, New York, and was a Rhodes Scholar. The views expressed here are those of the author and do not reflect the opinion of the Bank. 

The post When Computers Create Code, Who Owns It Is a Question Worth Billions appeared first on Artificial Intelligence.