Data Trends Series: AI Governance Update
Ville (Will) Kulmala ヴィッレ クルマラ
Data Strategy and Enterprise Technology Leader | Professor - Global Data Management | President of DAMA Singapore - Data Tuesday Tokyo (Data Management) | Is Your Data an Asset or a Risk - Know What You Own
Artificial Intelligence (AI) governance refers to the guardrails that ensure AI tools and systems are and remain safe and ethical. It establishes the frameworks, rules and standards that direct AI research, development and application to ensure safety, fairness and respect for human rights. (1)
AI Governance has been the hot potato in the Data Management area in recent months. There are too many developments and no one conclusion on this topic.
Views vary depending on whom you ask.
?
Here are the top 3 findings:
1) AI regulation is absurd or well needed, depending whom you ask
?
Benedict Evans (2) conclude that AI Regulation is the most boring thing in tech.
His readers are among the “magnificent seven” technology companies, yet nobody in the audience is interested in the topic.
?
His point is to stop regulating future developments, as the technology is too new to understand how it will shape society. Therefore, all AI’s ..
“.. create different problems, with different questions and consequences that require different kinds of expertise."
Another opposing view for regulation was raised by Ian Bremmer earlier this year on the risk of leaving AI ungoverned, consequences would be ..
“.. AI-generated disinformation and proliferation through state-backed large-language models and advanced applications for intelligence and national security.” (3).
In industries such as Financial Services, where transparency is required, the obscurity of AI platforms is a risk. Therefore, regulatory bodies are actively on the case to find the right approach for responsible AI.
It is expected that AI data would be vulnerable to breaches and thus regulation would follow in a similar fashion.
2) The New Gold Rush - Hoarding Data for AI Supremacy
Whereas the internet and crypto space were like the Wild West at the beginning, now AI is on a similar curve.
?
AI companies lack data and as The New York Times recent report (4) concludes, leading companies have been hoarding the internet for data without any consideration, only to train their large language models.
The resulting systems are impressive and have brought about ChatGPT and thousands of other AI companies (5).
?
Due to the extremely competitive state of current AI development, these companies have probably knowingly decided to kick the can down the road.
More legal showdowns are expected to arise in the next period regarding usage policies and content ownership which have been used to train popular AI models.
Asking ChatGPT about this, the AI admit that it is used for commercial purposes...
.. and even can make entertainment out of fair use copyright policy.
?
Data is the fuel of AI and therefore data management is at the center of AI Governance.
All companies are now de facto technology companies, AI is a layer to be more productive with existing high quality data - to do more with less.
?
3) No easy solutions on how to govern AI
?
According to the IMF (6), the AI power paradox is that AI can’t be governed like any previous technology, because it’s unlike any previous technology. It has unique characteristics which make it progressively harder to manage.
Data Privacy regulation is a good example of a technology regulation. As a result, companies are managing private data better and more securely.
This has led to many companies being fined billions due to data privacy breaches. Consequently, companies have invested in sufficient data management and governance practices.
An example of AI governance is the risk management led approach from California (7), which applies risk assessment to the usage or procurement of Gen AI solutions in the government.
The challenge, as always with regulation, is how to balance the speed of innovation with regulation.
The current evolution speed of AI is too fast for regulation.
AI models are learning constantly, while doing so, they create what is called “synthetic data”, which is information that is artificially generated from the existing learning data.
Even if the old training data is removed, models can't unsee the learnings.?
All this means is that it is complex and difficult to fully control data that is used and generated by AI and its underlying models.?
?
领英推荐
Asian perspective?
Similar to EVs, mobile, or e-commerce, China has built a competitive AI sector (14). India, too, has large pool of AI talent.
China has established a good amount of governance around AI technology. Some regulations are stringent centrally managed and some are done well and ahead of their time, such as the TikTok usage ban for teenagers (8), which has been in place for more than a year already.
?
Other countries in the Asia Pacific are pushing ahead as well, but maturity and implementation are still needed, even in data governance.
By 2028, 92% of workers across Asia-Pacific expect to use AI in their daily work (9) yet the current state of basic data and digital literacy is at most 29% in APJ countries. There is a big knowledge gap to bridge.
The borderless nature of AI means that the most advanced and powerful AI can be centrally hosted and can deliver AI capabilities in the local language and cultural context fluently.
This would mean, for example, that Gen AI which is hosted by a company based in Asia shall understand the local context and speak the local dialect perfectly to provide service at a local store in Mexico or Switzerland or vice versa.
AI brings the next evolution of globalisation by understanding customer needs in an even more comprehensive and hyperlocal way. Many traditional jobs will be transformed.
Countries will increase trade and technology barriers through regulation, such as AI governance.
Why??
One key reasoning would be to maintain welfare and jobs for the population, as the IMF suggests in their paper (10).
As AI continues to advance, the story of its governance is still being written, especially in Asia.
??????????????????????????????????????????????????????????????
Takeaways and Recommendations
?
1. Include AI and LLM models to data strategy
Takeaway: Start small and continuously align policies and outcomes in data strategy.
?
2. Establish Responsible AI role or a steering committee
Takeaway: Proactively participate in the local regulatory correspondence with the regulator. Adjust policies accordingly. Increase organisational data literacy.
?
3. Align data for the purpose and use case?
Takeaway: A best responsible AI practice for data management is the DAMA Data Handling approach related to people, processes, & technology. It covers AI data-related ethical perspectives. An adjustment is needed to cover all areas of AI governance. * (11).
?
?
Join us at:
Data Tuesday Kuala Lumpur 16th April
Data Tuesday Singapore 23rd April
Data Tuesday Tokyo 7th May
--?
Partner events :
Chief Data and Artificial Intelligence Officer - CDAO Singapore 2024
apidays Singapore 2024
---
References:?
(11)https://d3lut3gzcpx87s.cloudfront.net/image_encoded/aHR0cHM6Ly9zaWxrc3RhcnQuczMuYW1hem9uYXdzLmNvbS81ZWZkZjk5NTg5M2FhYTA0YjgwNzcyMDIuanBn/x (We await for DAMA to release updated framework which cover AI scope.)
Founder & CEO XdotO | Generative AI | Keynote Speaker | Masterclasses | Storyteller | AI Coach | ISB | IIM | Boards and CXOs | Enterprise Strategist@MINFY | Ex-CIO GE, Max Life, Jubilant, SMFG, SBI Card
7 个月Drafting and enforcing regulations are a huge challenge for both harnessing AI And curbing it's misuse. Very informative article. Loved the joke on BBC