Over the past year and a half, we’ve seen the audience for AI extend from IT experts to almost anyone with access to the internet.
Thanks to Large Language Models online users can now interact with AI systems, whether they can code or not.
In fact, a big part of the reason why Generative AI has been such a game-changer is because it has eliminated the need for specialised AI knowledge to experience the benefits of the technology. In short, the opportunity to democratise AI has never been greater.
And with AI increasingly in the hands of African innovators, we expect to see faster and more profound progress in nearly every field of human endeavour.
However, for Africa to meet this opportunity and truly democratise AI, there is still much work to be done around accelerating access to internet connectivity and growing digital literacy – which is why this has been such a central focus of Microsoft’s investments in the continent over the past 30 years.
We are also keenly aware that the challenges ahead extend beyond the need for greater investment in essential infrastructure and capabilities.
In fact, many public and private organisations in Africa view the risk of new safety and regulatory requirements as one of the biggest stumbling blocks to wider implementation of the technology.
It’s clear that to fast-track AI, leaders will need to work together to advance its governance more quickly.
To support these efforts, we compiled a report titled “Governing AI: A Blueprint for the Future” which was spearheaded by Microsoft’s Vice Chair and President, Brad Smith. This has since been reworked to suit the needs of different countries and regions and earlier this year we launched, “AI in Africa: Meeting the Opportunity”.
This whitepaper explores the five key focus areas that can contribute positively to the work ahead.
1. Implement and build upon new government-led AI safety frameworks
When it comes to using AI safely, one of the most effective ways to accelerate progress is to build on existing governmental frameworks.
Several African countries have already begun to formulate their own legal and policy frameworks and are helping to lead discussions around AI policy and strategy development on a regional, continental, and global scale, offering valuable insights for other countries looking to do the same.
While at different stages of implementation, they all are looking to find balance between the need to create guardrails for the new technology while at the same time wanting to help a nascent industry grow, innovate, and adopt these new and emerging technologies.
The African Union (AU) continues to convene experts from across the continent and this year published a policy draft containing a comprehensive continental strategy for AI regulations for African countries.
Their coordinated approach aims to consider the “responsible, safe and beneficial use” of the technology for all Africans.
Once adopted, this framework would help countries that lack AI policies or regulations to create their own national strategies and would also urge those that have them to revise and harmonise their policies with the AU’s.
The draft proposes guidelines for codes and practices that are tailored to different sectors, standards and certification bodies to evaluate and compare AI systems, regulatory sandboxes for secure testing of AI, and the creation of national AI councils to supervise and track the responsible use of AI.
2. Require safety brakes for AI systems that control critical infrastructure
While most potential AI scenarios do not pose significant risks, it’s going to be increasingly important to identify those high-risk situations that will require ‘safety brakes’.
This is particularly relevant when it comes to systems that manage or control critical infrastructure such as electricity grids, water systems, traffic systems or emergency responses. These brakes ensure systems can be quickly controlled or stopped by humans if necessary.
One way for governments to begin developing this safety mechanism is by defining the class of high-risk AI systems that are being deployed to control critical infrastructure and then requiring developers to build and embed such added layers of security in the form of ‘safety brakes’.
From there, operators can rigorously test and monitor these high-risk systems, making certain that they can avoid unintended consequences and remain under human control.
This approach highlights the need to create AI systems that can recognise, prevent, and stop unwanted behaviours, following the best standards in human-computer interaction.
3. Develop a broader legal and regulatory framework based on the technology architecture for AI
To address AI’s legal and regulatory challenges, a framework mirroring AI’s technology architecture is needed, focusing on the three layers of the tech stack, with different obligations for the level of applications and the layers beneath, which are the AI foundational models and the infrastructure.
The law will also need to place various regulatory responsibilities upon different actors based upon their role in managing the different aspects of the AI technology.
Laws should apply existing protections to AI applications, ensuring safety and rights without necessitating new regulations, as current laws can often be adapted.
For foundational AI models and their deployment, however, new regulatory approaches are necessary, especially for models with significant capabilities.
This could include a multitiered licensing regime, enforced by the government, for the development and deployment of highly capable AI models, emphasising safety, security, and international cooperation.
One of the initial challenges will be to define which AI models should be subject to this level of regulation. It will be necessary for leading AI developers to share specialised knowledge about advanced AI models to help governments define the regulatory threshold.
They can then outline the requirements that must be met to obtain a license to develop or deploy a highly capable AI model.
An effective licensing regime will help to ensure that we maintain control over our electricity grid and other AI-operated infrastructure.
Additionally, AI datacentres, critical to the operation of these advanced AI models, should be mandated to meet specific requirements to ensure the responsible development and deployment of these technologies.
4. Promote transparency and ensure academic and public access to AI
A key aspect of AI policy that will require serious discussion in the coming months and years is the balance and tension between security and transparency.
For example, some experts think that keeping AI model weights (which are parts of a model that are crucial for a model’s abilities) secret will be necessary for security as this might be required to safeguard vital national security and public safety interests.
However, in many other cases, transparency will be important to improve the understanding of security needs and develop best practices.
This is why it’s important to think through how one can provide different types of transparency in different circumstances.
Transparency reports can play an important role in driving corporate accountability and empowering the public to understand AI systems, including where and how they are being used.
Ultimately, the public needs to be informed when content has been created by AI. This also applies to original content that has been altered using AI.
Another aspect that adds to transparency is to provide broad access to AI resources for academic research and the nonprofit community.
Academic research has and will continue to play an important role in societal innovation and the nonprofit sector will be critical to ensure that AI technology remains accessible and accountable.
5. Pursue new public-private partnerships to use AI as an effective tool to address the inevitable societal challenges that come with new technology
AI is a powerful tool with immense potential for good. Like with other technologies, however, there are some who will try use it as a weapon. Fortunately, the technology can also be harnessed to fight against the abuse of AI and to address societal challenges.
Public and private partnerships between governments, companies and NGOs will be needed to drive progress in this and other key areas, from skills development to sustainability efforts.
We’ve seen the effectiveness of this approach in countries like Nigeria where we’ve partnered with the United Nations Development Programme to co-convene the AI for Development Reference Group, a multi-stakeholder and interdisciplinary collaboration tasked with helping shape the country’s AI agenda.
In Kenya, we launched a Responsible AI series with Bowmans, Strathmore University, and other stakeholders to discuss AI policy, regulatory frameworks and governance within the country’s AI ecosystem.
And in South Africa, Microsoft’s partnerships with key players across the public and private sectors are reshaping public service delivery and addressing multifaceted business and societal challenges in areas such as health care system optimisation.
Collaboration with the Desmond Tutu Health Foundation (DTHF) at the start of COVID-19, for example, contributed to crucial COVID-19 vaccine trials.
A system harnessing Microsoft Dynamics 365 and Microsoft Azure Cognitive Services was implemented to help the DTHF manage various healthcare services and clinical trials, providing the Foundation with real-time access to data to improve efficiencies and reporting while also enabling the automated booking and scheduling of participants.
Ultimately, the DTHF was able to enrol and manage study participants much faster and more efficiently.
These are of course a few examples with many more initiatives taking place across Africa.
While we certainly don’t have all the answers to the questions that this new AI era brings, we believe that by working with stakeholders across the continent, we can help shape a future where AI is a tool that benefits everyone.
Grounded in responsible regulation and collaborative partnerships, Africa can fully realise the opportunities presented by a future with AI.
Author
Comments 1