Headlines in Tech 27 Nov

Headlines in Tech of the week

US – EU Trade and Technology Council initiatives indicate closer co-operation on approach to trade, technology and innovation in line with democratic principles and universal human rights

…this of course includes co-operating on new emerging technologies. It is likely to impact some developed third countries in particular (which sadly does include the UK, yes) because once certain standards are set for implementation in the US and EU, businesses would apply them to other similar markets:

AI – which includes a roadmap. This includes AI standards that articulate requirements, specifications, test methodologies, or guidelines relating to trustworthy characteristics to help ensure that AI technologies and systems meet critical objectives (e.g., functionality, interoperability) and performance characteristics (e.g., accuracy, reliability, and safety).

Pilot project on privacy enhancing technologies and synthetic data in health and medicine

Establishing an expert task force on quantum information science and technology to reduce barriers to research collaboration, and engage on standards, intellectual property, and export control-related issues

Developing joint recommendations for government-funded implementation of EV charging infrastructure

Increasing standards cooperation on additive manufacturing, recycling of plastics, and digital identity

Launching new workstreams on post-quantum encryption and international cybersecurity standards

Artificial Intelligence

OpenAI launches ChatGPT – causes stir. Here is why

…if you scroll right to the bottom. Why has there been so much hype? How does it differ from AI already on the market? What are the issues with it? All addressed below.

Google licences AI mammography models and personalised risk assessment to medtech company iCAD

…this should improve breast cancer detection and short term personal cancer risk. It will also alleviate burdens on shortage of staff able to review mammography and anxious wait times for patients.

Researchers say Alphafold’s predictions on protein structures very much require verification with experiments

…it says some AlphaFold predictions match experimental maps closely, but most differ on a global scale…

Researchers claim new AI model can predict risk of death from heart attack of a stroke within the next 10 years based on a single X-Ray

… the AI model used more than 147,000 chest X-rays from more than 40,000 patients who participated in the Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial. Researchers say there was a significant link between the risk predicted by their algorithm and the risk of experiencing a major cardiac event. It can be used to identify patients who might benefit from blood thinning medication. A randomised trial is necessary though to confirm AI’s accuracy.

Amazon’s Alexa can generate animated bedtime stories for children

…Children can specify the type of stories (eg: theme – space exploration /name of character, colour themes and overall tones, such as silly or happy), enabling Alexa to generate an original story. It can generate different stories based on the exact same prompts as well as children are want to do. The response of many might be – but it’s really the parents/ guardian’s job. Plus query how healthy it is to be looking at a screen just before bed time. Amazon says it helps the child be creative because they can co-create the story. My own ask would be a provision of a parental tool to somehow surreptitiously include a particular message (eg. importance of brushing teeth carefully – an issue I am having at home right now), to persuade the child to listen to what the parent says!

Artificial intelligence computer maker Cerebras Systems partners with cloud computing service provider Cirrascale to offer pay per model compute time to speed the use of very large language models

…this would help businesses with the cost of renting cloud capacity or lease machines to carry out the work. Prices range from $2,500 dollars to train a 1.3-billion-parameter model of GPT3 in 10 hours to $2.5 million to train the 70-billion-parameter version in 85 days (GPT3 is a language model that uses deep learning to produce human-like text following a prompt).

Amazon publishes service cards for users to help them use Amazon’s AI Services responsibly

…there are three at present, Amazon Rekognition — Face Matching, Amazon Textract — AnalyzeID, and Amazon Transcribe — Batch (English-US). For example, for the use of Amazon Rekognition (which enables customers to understand how well two faces match, or identify whether a face is likely to be one of the faces within the customer’s face database) the service card raises matters such as use cases, performance expectation (eg. degradation of performance where the image is blurry), effect of variables (eg. hairstyle), fairness and bias, explainability, privacy and security (eg. Face vectors are never included in the output returned by the service, Inputs and outputs are never shared between customers), transparency and governance, and best practices.

BigTech/ Data / Platforms

Algorithms

Real Women in Trucking says Meta’s algorithms discriminates against women and older people in Facebook job listing

…the complaint says that “even when employers directed Facebook to send their job ads to people of all genders and ages, Facebook delivered the ads to Facebook users who are over 99% male and 99% younger than 55 years old”. Certain job adverts are exposed to mainly men and young people (trucking, construction, firefighting, manufacturing, warehouses, mechanics, engineering etc), and others to mainly women (housekeeping, food service, administrative assistants, paralegals, nursing, home care, social services, and child-care). This follows similar complaints made against Amazon and T-Mobile for placing ads on Facebook which discriminated against older candidates. Those though are now settled.

Business

Business in hardware is hard, demonstrates Amazon as it axes hardware related workforce

…Workforce dealing with Alexa voice assistant, Kindle e-reader and Halo health tracking have been laid off. That said it is still pursuing the acquisition of vacuum cleaning robot Roomba. I thought Amazon was all in on smart home business, so I was a bit surprised.

Facebook will remove news from platform (in US) if it is compelled to pay news group – per the Journalism Competition and Preservation Act (US bill)

…Meta does have in place deals with news groups, but do not wish payments to be mandated. It also says that it helps bring new users to the news sites themselves. The issue though may be local news businesses that may suffer as a result of content on the Facebook platform.

Competition

Digital Cartel? Rent-setting software company RealPage under investigation by US Department of Justice

…RealPage’s software YieldStar uses private rent information received from landlords to come up with rental rate recommendations to its users, which then indirectly gives landlords (including major ones) access to competitor pricing. There is also concern over RealPage user group, comprised of apartment managers which may use the channel to coordinate pricing.

Amazon settling with the EU Commission to align with the Digital Markets Act

…the antitrust complaint though was initiated by the EU Commission well before the DMA came into play. Under the deal, Amazon will ensure that rival products will be equally visible on its eCommerce marketplace (ie: in order to comply with the ban on self-preferencing), it will provide users with alternatives where the speed of delivery is not of essence, and sellers using Prime membership will now be allowed to use a logistics company other than Amazon.

The deal is though not yet set in stone. The DMA was enacted to ensure fair and open digital markets. The EU Commission explains: When a gatekeeper engages in practices, such as favoring their own services or preventing business users of their services from reaching consumers, this can prevent competition, leading to less innovation, lower quality and higher prices. When a gatekeeper engages in unfair practices, such as imposing unfair access conditions to their app store or preventing installation of applications from other sources, consumers are likely to pay more or are effectively deprived of the benefits that alternative services might have brought.

Content Moderation

UK to scrap platformers’ duty to scrap legal but harmful content

…following fierce lobbying from BigTech and free speech advocates. But the Online Safety Bill will still remain strict:

Requirement to be transparent on content moderation policies
Clarification on platforms’ age verification process – to protect children from bullying and pornographic content. Any online child abuse to be reported to UK National Crime Agency
Requirement to publish risk assessment
Addition of new offences: Encouragement of self-harm or suicide, uploading of non-consensual images such as deepfake porn

Note that when it comes to protecting children, the US and the UK may have different considerations. In the US, LGBTQ+ books are banned in schools, and some healthcare providers that are treating trans children are reportedly falsely labelled as grooming – in that climate, it is vital that online platforms do not over content moderate. Certainly this is the position of the opponents of s.3663 the Kids Online Safety Act which is presently being considered. That type of consideration may be less acute in the UK. Children’s privacy is also an issue as it compels platforms to allow parents supervise minors, including 15 and 16 years who might be considered to have their own independent right or privacy.

Meanwhile Elon Musk is complaining that Apple has stopped advertising on Twitter (Does Apple hate free speech in America? He rhetorically asked) and that it had threatened to ban Twitter on the Apple App Store – but won’t tell him why (although following tete-a-tete with Apple CEO Tim Cook at Apple HQ, was resolved as apparently a “misunderstanding” –, in practice Twitter needs Apple (because it controls which App can feature in the App Store on iPhones) and Apple is not in the position to cut off Twitter without a really good reason (its control over the smartphone app space is being scrutinised)).

Musk has also posted a poll on whether “Apple should publish all censorship action it has taken that affect its customers” and commented “Secret Suppression of free speech by Apple. Customers were never told. What the hell is going on here?”. The Online Safety Bill would seemingly address some of Musk’s concerns [note though Apple does publish App Store guidelines, which include apps duty to block abusive users, report offensive content, filter objectionable material from being posted. The latter one seems to me to be more like the duty to remove legal but harmful content].

EU threatens Musk that Twitter may be shut down if content moderation is not put right

…it must stop arbitrarily reinstating banned users, allowing spread of disinformation, and comply with the Digital Services Act (this is EU law which objective is to ensure that platforms remove illegal content). It must agree to an extensive independent audit, the EU Commissioner added. In any event, as a result of lack of content moderation in part, advertisers have been fleeing Twitter. That should present ample incentive for Twitter to content moderate. This week Musk has had to ban friend, rapper and US Presidential hopeful Ye (formerly Kanye West) from Twitter for tweeting a symbol which combined the swastika with the Star of David – on the ground of inciting violence (which is unlawful). ThisTwitter is now incentivising advertisers to spend on their website, giving advertisers additional impressions if they spend a certain amount with Twitter.

The US, on the other hand, is concerned about foreign ownership of the US social media platform, which includes a stake held by Prince Alwaleed bin Talal bin Abdulaziz of Saudi Arabia. President Biden is said to float the need to look at Musk’s “co-operation” with other countries. That would involve a look inside “co-operation” he has with the Chinese government in relation to the Tesla business too.

Meta’s own Oversight Board says Facebook content moderation policy is unfair because it’s more lenient on VIPs

…allowing influencers’ offensive content to remain on the system for longer. The special VIP review program is called “cross-check,” or XCheck. Meta spokesperson explained “We built the cross-check system to prevent potential over-enforcement (when we take action on content or accounts that don’t actually violate our policies) and to double-check cases where there could be a higher risk for a mistake or when the potential impact of a mistake is especially severe”.

Privacy

Meta fined €265million by Irish Data Protection Commission because Facebook’s features lacked Data Protection by Design and Default in breach of GDPR

…the scope of the inquiry concerned an examination and assessment of Facebook Search, Facebook Messenger Contact Importer and Instagram Contact Importer tools. Personal data of more than 500 million Facebook users across more than 100 countries were allowed thereby to be scraped and posted on a hacking forum. The data included names, locations, birthdays, passwords, phone numbers.

Austrian privacy activist Max Schrems complaint succeeds again – European Data Protection Board says Meta must get users’ explicit consent before tracking in compliance with the GDPR

…but decision will be appealed no doubt.

Class suit filed against Tinder for storing collected biometric data

…Tinder collects these to ensure that the person who is using the service looks like the photos they are uploading. Complaint says Tinder has failed to state how the biometric data will be used and how the data will be destroyed.

Facebook Dating says it will share third party AI scanning tool (Yoti – UK based company) to verify the age of its users. Yoti says it will delete the images immediately once verified. Facebook offers an alternative method if you don’t want your face to be scanned, by proving age by submitting information on passports and driving licences.

In a class suit, TikTok is alleged to have collected massive amounts of data by tracking users’ activities on the internet

…the complainant says the in-app browser inserts JavaScript code into the websites visited by TikTok users to track every detail about TikTok users’ website activity, and then to generate revenue. The cause of action is the breach of wiretap laws and consumer protection laws.

Blockchain/Crypto

Global supply blockchain platform and IBM / Maersk JV TradeLens to shuttter

…shipping involves a lot of paperwork. Documentation entailed differ from country to country – bills of lading, customs, insurance differ etc. There appeared to be huge scope for improving efficiency (including trackability) and security by digitising and managing all information on a blockchain. TradeLens invested efforts into establishing data standard and getting user buy in – but failed to achieve commercial viability.

EVs/AVs

South African miner Sibanye-Stillwater receives permission to mine and refine lithium in Finland

…Sibanye-Stillwater is one of the largest producers of platinum and palladium. The latest development follows the granting of a planning permission to build a refinery in Tees Valley in the UK (by Trafigura backed Green Lithium).

The aim is to reduce reliance on China, which controls 60% of global capacity to process lithium raw materials into battery-grade chemicals (the hard rock containing lithium is reported to be mainly in Australia, or else lithium containing brines can be extracted in Chile and Argentina).

Mercedes offer better acceleration feature for an annual fee of $1200

… the feature will give a noticeable improvement in acceleration of 0.8 to 1.0 seconds (0-60 MPH). This type of subscription based value add offering follows the launching of a heated seats service by BMW for a fee of $18/month.

Gaming/Metaverse/NFTs

EU’s €400m spend on metaverse criticised

…last year, the EU Commission launched the Global Gateway, the “new European Strategy to boost smart, clean and secure links in digital, energy and transport and strengthen health, education and research systems across the world”. It is now reported that the Commission has spent nearly €400m to engage youngsters with the EU and its activities. The digital platform is reportedly “a dedicated metaverse where one can explore what the Global Gateway means through a series of ‘hero’ stories in a virtual environment”. WTF, apparently said one critic.

Robotics

US Food and Drug Administration OKs Bionaut Labs to accelerate through to clinical trials an implantable device which delivers drugs to specific parts of the brain using magnetic forces

…they are now raising funding. Animal studies have now been done. It will trial use to treat a type of brain tumour and a rare paediatric neurological disorder Dandy-Walker Syndrome but eventually to other more common brain diseases such as Parkinson’s, Huntington’s, Alzheimer’s and stroke. Local drug delivery has a better chance to reduce side effects. Drug delivery in the brain is particularly difficult due to the blood brain barrier, a barrier our bodies have evolved to have to protect harmful substances reaching the brain – enabling us to control what contents in the blood can reach the brain.

Musk’s brain/computer interface business Neuralink investigated for its treatment of animals

…Neuralink employees said that the business sacrifices more animals than necessary, and experimentation isn’t carried out as carefully as it should be. This is a competitive area, and Musk is pushing progress – he is reported to have said to employees to imagine that a bomb was strapped to their heads to achieve results quicker. Its products are aimed at helping patients with spinal injuries improve their functions.

Satellites/ Space

UK to trial SpaceX’s Starlink service to provide fast internet across the country

…the government aims to tackle the digital divide thereby. Starlink was chosen because it was available and cheaper, but the part UK owned, UK base satellite provider OneWeb may also have the chance to trial in the future.

In the Spotlight

OpenAI launches ChatGPT – causes stir. Here is why

… it can interact with you in a conversational way. OpenAI says ” The dialogue format makes it possible for ChatGPT to answer follow-up questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests”. It can correct codes as well. So you ask it to do something, like write a poem about visiting the seaside before Christmas. It then generates a poem. You can then say, “do another one”, or be more specific, “do another one with a heartwarming tone”. You can try it out for free. Or you can combine different AI models – so get ChatGPT to come up with some ideas on decorating a living room, and insert the answer to another Generative AI model, like this guy did here.

Content moderation

It is programmed not to answer inappropriate questions, or instructions that may cause harm (sexual, hateful, violent, or promotes self-harm), or exhibit biased behaviour – through the use of the Moderation API to warn or block certain types of unsafe content, but it does not expect it to be perfect.

Copyright issues

Generative AI is most certainly trained on copyright protected content, in a large part scraped from (ie: reproduced/copied from) the web which comprises work of artists, photographers, writers and well, creators generally. The system is now imperilling their livelihood. This is a problem. Note that the Berne Convention for the Protection of Literary and Artistic Works – to which 181 of 195 countries are a member, stipulates at Article 9.2 that “it shall be a matter for legislation in the countries of the Union to permit the reproduction of such works in certain special cases, provided that such reproduction does not conflict with a normal exploitation of the work and does not unreasonably prejudice the legitimate interests of the author”. Therefore, the perennial copyright issues apply such as whether the scraping the web for the creation of generative AI involve copyright infringement. What about the output of the AI – would it attract copyright, and if so who would own it? If that depends on the degree of human involvement, to what extent should the human be involved (would typing a prompt – such as the creation of a seaside Christmas poem suffice)? It may depend on the purpose for which the AI is created or used (eg. educational purposes). The US Supreme Court is deciding a case about the legality of Andy Warhol’s artworks of Prince, which involved copying protected works of a photographer, of Prince. The Warhol side is saying it was within the US fair use exception, on the grounds that the work is “transformative”. It may have implications for Generative AI.

There is a proposed class action in the US against OpenAI and Microsoft in relation to Microsoft owned GitHub’s generative AI powered coding assistant Copilot. Copilot is a service that suggests lines of code and functions in real-time, which was built using code that have licences attached to them, such as the requirement to credit the original programmer. One of the issues would be the same as the Prince case, namely whether such use fall within the fair use doctrine. Note that in the EU, exceptions to copyright is based on an exhaustive list of exceptions and limitations, rather than a more general fair use system deployed in the US.

How it is trained

OpenAI explains that it trained this model using Reinforcement Learning from Human Feedback (RLHF), which fine tunes the model depending on how humans rated the output of AI (called a reward model – then eventually the AI can learn to rank how good the model is). However, the problem is that the output can be completely incorrect yet framed in a very authoritative, convincing manner. The error rate was high enough to compel Stack Overflow, which is a question and answer site for programmers, to ban users from sharing responses from ChatGPT3.

Microsoft involvement

The ChatGPT3 uses Azure AI supercomputing infrastructure (OpenAI entered into an exclusive computing partnership with Microsoft to build new Azure AI supercomputing technologies in 2019). OpenAI is basically Microsoft’s bet, OpenAI gets the compute power free from Microsoft, and Microsoft gets to embed new technologies into its offerings (for example it recently released a Designer app with Generative image generating AI Dall-E (the product of OpenAI) in its interface). The compute power needed is so massive though, OpenAI is saying that it is not possible to offer the use for free for forever.

Note

Why is there so much hype over ChatGPT3? The interesting point to note is that ChatGPT3 does not represent that much of a development technologically in the purest sense. It is built on a slightly upgraded version of GPT3 (third version generation pre-trained transformer, ie: language model version 3. A transformer is a type of neural network that learns context, so it can extract meaning from data), which has been around for over two years. The key is the change in the interface – it’s in a conversational form, meaning much more user friendly.

What’s the difference between Generative AI and the more traditional type of AI?: In basic terms, the classic concept of AI is to find patterns and predict something. The answer is based on statistical probability. An example that is often used is an AI which is able to discern an image of a bus from a car. The AI has been fed so many images of vehicles and has been trained to discern a bus over other things, such as a car – by us! Many will have experience having to click on images of a bus, or a traffic light or whatever to prove you are not a robot. Same with hashtags you put on social media posts (eg. #schoolbus). The AI then learns the shapes of buses and cars, and when it is presented with an image, it will carry out a statistical analysis and say, the body is long, so likely this is a bus, not a car. In contrast, Generative AI can create something new. So you can ask a Generative AI, for example, what will a combination of a bus and a car look like? The AI may never have seen anything like that before, but it can create new material based on what it knows.