#6: Ensuring information quality is essential for every company in the Age of AI
Getting information quality right must be a board level responsibility
Hello from Singapore. The city has very much regained its verve in the past year after two years of Covid-based lockdowns and border closures. The rest of the world has so much to learn from this incredible country and many of these lessons are not the ones being articulated by British and American libertarians. More of that to come…
Thanks again to everybody who has subscribed and shared to this Substack and to those who have shared their thoughts via email too. You can reach me on david@artemonstrategy.com and I’d love to hear from you.
Our website is www.artemonstrategy.com
A bit of a deeper dive this week into a crucial issue as we enter the age of AI. Namely, the importance of ensuring information quality and how this must be the board level responsibility of all companies utilising AI, rather than simply tech firms or technologists.
Before we kick off, it’s also well worth reading this important piece from the FT about how leaders “can put organisations at risk when they fail to plan for the worst case scenario.” At Artemon, we regard scenario planning as fundamental to business success. Please get in touch if you want to chat about how we could work with your organisation.
Ensuring information quality in AI has to be a board level issue
The world is rightly being wowed by the potential of generative AI. This week, Bill Gates correctly argued that the transformative power of AI is:
as fundamental as the creation of the microprocessor, the personal computer, the Internet, and the mobile phone. It will change the way people work, learn, travel, get health care, and communicate with each other. Entire industries will reorient around it. Businesses will distinguish themselves by how well they use it.
Chart via Grand View Research
The potential for AI to drive positive transformation across industries is undeniable. It’s already driving huge advances in healthcare and leading to great strides in office productivity tools. There’s not an industry that isn’t considering how to utilise generative AI in their processes. This includes industries from wine making to football. I even saw an advert in Singapore this week for an armchair that claimed to be powered by AI.
When AI transforms whole industries it moves from being a “tech company issue” to being a board level issue for all businesses.
There is, of course, another side to AI that society must be ready to guard against. This isn’t the dystopian concept that an out of control AI could be destructive to humanity, but rather the known issue that leaps forward in AI technology can also accelerate known issues with technology - spam, privacy, targeting of ads, bad content and fairness related concerns. Possibly the most important concern is the quality of information, including the potential misuse of AI misinformation and state-level misinformation, which might suddenly become more realistic, more plausible and more targeted.
In an age of generative AI, it will be the responsibility of every company that utilises generative AI to ensure that the information is of the highest possible quality. In the AI eco-system ensuring information quality will be the responsibility of all AI-based companies, not just tech firms. Within a company ensuring information quality will become a board level responsibility, rather than compartmentalised as the responsibility of technologists.
It will be the responsibility of all boards who utilise AI to ensure that the information that the organisation operates on will be of the highest possible quality.
Information quality will be crucial for all companies
At its most innocent, companies in all sectors using generative AI need to avoid some of the basic factual errors that have been seen in otherwise impressive early large language models and chatbots. Any company using something like ChatGPT needs to avoid exposure to factual inaccuracies inaccuracies ranging from Hobbes believing in the separation of powers (he believed in a Leviathan state so absolutely did not) to York racecourse having an uphill finish (it doesn’t). GPT4 might have shown increased accuracy to its predecessor and even passed the US bar exam but large language models continue to be plagued with inaccuracies and misguided information. Existing models are also prone to what have been termed “hallucinations”, where they effectively make up an answer to a question. Such mistakes might be expected in early large language models which are effectively part of an ongoing experimentation, but they will be less forgiven if companies across sectors integrate such AI into their core services and users expect an accurate “single answer”.
Low information quality could also blunt the potential for AI across sectors. Potential inaccuracies could limit the ability for AI in banking, for example. The creation of legal content through AI, such as is already been practiced by Allen & Overy and their Harvey platform, needs information quality to be guaranteed. Many white collar occupations that could see major changes through generative AI will need assurance of information quality if AI is going to have maximum beneficial impact.
AI could accelerate disinformation threats
The nature of the advances in generative AI mean that disinformation campaigns might be a perilous underside of the new technologies. The ability to believably write documents (or poems, or song lyrics, or screenplays) in the style of well known people was one of the most newsworthy features of Chat-GPT when it burst onto the scene late last year. What is to stop bad actors using such a technology for nefarious purposes? AI’s ability to accurately impersonate somebody is risk enough, but the ability to target and personalise this mimicry potentially increases the danger even more at a time of state-sponsored disinformation.
This is, of course, acknowledged by AI developers themselves. As Open AI put it in their technical analysis of GPT4:
GPT-4 can generate plausibly realistic and targeted content, including news articles, tweets, dialogue, and emails…. similar capabilities could be misused to exploit individuals… Based on our general capability evaluations, we expect GPT-4 to be better than GPT-3 at producing realistic, targeted content. As such, there is risk of GPT-4 being used for generating content that is intended to mislead.
Building guardrails against such misuse is obviously crucial not just for tech firms but for all companies engaged in AI.
The 2022 WEF Global Risks report identified “cross border cyber attacks and misinformation” as one of the most important global risks and also “ among the areas with the least ‘established’ or ‘effective’ international risk mitigation efforts.” The increasingly polarised geopolitical climate has led to a growth in state-sponsored disinformation campaigns and cyber attacks. The 2022 Microsoft Security report (from which the image below was taken) found a worrying increase in state sponsored attacks, which have become “increasingly sophisticated” and “more brazen and aggressive as geopolitical relationships have broken down.” and Freedom House has found state sponsored disinformation campaigns to be increasingly sophisticated and effective.
An analysis by Newsguard of the most recent large language models found most recent models to be “more susceptible to generating misinformation — and more convincing in its ability to do so”. It suggests they are already “skilled at generating responses in the voice of a particular person or outlet, including, as NewsGuard asked among its queries, prominent misinformation purveyors, Russian state-controlled news outlets, and fringe conspiracy theorists.”
The Centre for Security and Emerging Technology at Georgetown also detailed the potential risks of the emerging technologies, when they suggested that:
Artificial intelligence (AI)… is poised to amplify disinformation campaigns… the use of AI in disinformation campaigns is not only plausible but already underway. Powered by computing, ML algorithms excel at harnessing data and finding patterns that are difficult for humans to observe. The data-rich environment of modern online existence creates a terrain ideally suited for ML techniques to precisely target individuals. Language generation capabilities and the tools that enable deepfakes are already capable of manufacturing viral disinformation at scale and empowering digital impersonation. The same technologies, paired with human operators, may soon enable social bots to mimic human online behavior and to troll humans with precisely tailored messages. These risks may be exacerbated by several trends: the blurring lines between foreign and domestic influence operations, the outsourcing of these operations to private companies that provide influence as a service, and the conflict over distinguishing harmful disinformation and protected speech.
A Stanford report also underlined how large “language models will be useful for propagandists and will likely transform online influence operations.” They could, if used improperly, effectively provide personalised, tailored propaganda at scale, with AI meaning that both video and audio deepfakes become much more realistic and much more concerning.
What ensuring information quality means for businesses
Companies across the economy stand to benefit from the potential of AI. But making the most of this requires boards of any company utilising AI to ensure information quality in a number of ways:
Every company looking to utilise AI needs to develop an AI strategy. AI that will transform industries can’t be considered merely a “bolt on” to existing offerings, nor should it be regarded as an issue purely for technologists. How to use AI will be fundamental to a company’s strategic direction.
Ensuring information quality should be an essential part of an AI strategy. It is the board’s responsibility to ensure that the information the organisation operates on is of the highest possible quality. Getting information quality wrong could have a damaging reputational and business impact. Any company using text-based generative AI will have to put their own safeguards in place to ensure information quality. This is a board and company level issue, not simply an issue for tech firms and CTOs.
Ensuring good, structured data should be fundamental to an AI strategy. It is a mistake to think of data purely in isolation. Companies should be thinking in terms of structured data. The same data can be included in both high quality and low quality information - it’s how that data is related to other data as part of a structure that dictates the quality of information.
Elevating high quality information will be as important as removing low quality information. Much of the discussion around tackling misinformation (or simply poor quality information) is focused on detecting and removing the poor quality information. But companies should also consider how to detect high quality information and amplify it. This would help to reintroduce authority within an information ecosystem that so often lacks it. The new generation of tech development if considered seriously could play a role in elevating authority within information.
Companies should develop scenarios and wargaming simulations about how their AI strategy might run into information quality and other challenges. Foresight capabilities, such as scenario planning and wargaming, will be crucial for companies looking to navigate the emerging environment of generative AI and helping them to both maximise opportunities and minimise potential bear traps.
Partnership between tech firms, business and government is essential to building high quality information. Effective partnerships will be a crucial building block to ensuring that AI is built on high quality information and is not exploited. This can fall into several different avenues of partnership. First, effective partnering between governments and tech firms will be essential to tackling more sophisticated misinformation. Second, tech firms should offer support and partnership to non-tech companies in navigating information quality challenges, potentially through a more formal organisation. Third, business, academia and civil society need to accelerate collaborative working to accompany the acceleration of generative AI and its associated challenges.
And finally…
Some other notable stories spotted this week:
Vinyl sales have overtaken sales of CDs for the first time since the 1980s. 84% of music listening is done via streaming though.
Generative AI and newspapers - could there be a looming copyright conflict?