For our August Twitter Chat on AI Ethics, we invited Maria Luciana Axente(@Maria Luciana Axente), program driver for Artificial Intelligence (AI) and AI for Good lead at PwC UK to discuss how we can use Artificial Intelligence for Good.
Mia Dand: Thank you so much for joining us today, Maria. Let’s start off with the basics. What does AI for Good mean?
Maria Luciana Axente: There are many definitions for AI for Good but the one we prefer is one used by United Nations (UN)- beneficial use of AI technologies to address the Sustainable Development Goals (SDGs), adopted by all 190 states, a universal call to action to address humanity’s biggest problems. SDGs represent humanity’s biggest challenges to end poverty, reduce inequality, protect the planet, clean and smart cities, and waters and ensure that all people enjoy peace and prosperity. All to be achieved by 2030. The SDGs could be grouped as economic, social and environmental, hence a subcategory of AI for Good is AI for Social Good, AI for Earth, Green AI, etc.
There are other taxonomies being used, mainly in defining the type of problem being addressed by AI like humanitarian issues, social issues or the output as a good and Responsible AI. Responsible AI defines application who have the right values embedded throughout the life cycle ( fairness, transparency, privacy, etc.) acknowledging the context and are governed in such a way that risks are mitigated and quality is achieved. It is important to define the right taxonomy used to provide clarity on what types of problems are being addressed to benefit from cross-sharing solutions, methodologies, and capabilities — like data, algorithms and standards.
MD: Completely agree on the need for using a clearly defined taxonomy so we can make faster progress in this space. What some examples of AI for Good?
MLA: Some big ones, in the current debate of climate crisis are the use of AI to address the biggest environmental challenges using AI. This report by Celine Herweijer and Ben Combes describe a range of use cases for AI for Earth My heart goes to using AI to tackle human trafficking especially 4children, or identifying children abuse cases. If we start solving what impacts the wellbeing of children, a lot of other problems would be solved! A big shout-out to AI for Good team at Element AI led by Julien Cornebise. The human rights application of AI again stands out. Using AI to protect every person in life-threatening danger is ESSENTIAL. But a lot of SDGs still await AI solutions! SDG 6: ‘Ensure access to water & sanitation for all’ has little attention while 700 million people affected by it. Share it with your network and let’s start addressing the issues. For a most up to date repository of #AIforGood solutions, using the #SDG taxonomy, visit the ITU website to learn more about the fantastic work they are doing.
MD: How can for-profit corporations balance AI for Good with business objectives?
MLA: ROI for AI is a challenge in general. It is important to consider the type of AI for Good problems being addressed, define the benefits (tangible & intangible) and allocate the right resources. Clarity of scope is essential to create value & impact.
Good news — the most important SDGs like health, education or smart cities benefit from funding, either public or private. For those use cases, a high level of transparency, accountability & governance is A MUST to manage large scale impact. In the UK, NHSX, a branch of NHS is doing great work on AI applications in health across its value chain, from diagnostics to optimization of operations. Check out the great work their AI lead, Jess Morley and team are doing.
However, there are many other #SDGs lacking funding and attention like poverty, reduce inequality or environmental ones. Unfortunately, addressing the SDGs is prioritized based on commercial value, not so much on the impact or scale of those issues.
MD: Which department or function in the company should lead AI for Good efforts?
MLA: I see two options, one is to create a separate AI for Good function like Element AI or Microsoft. Another option is to make it part of the central AI team like PwC UK.
The structure needs to be cognizant of the culture, organizational setup, and build on the strengths & internal opportunities. Leadership buy-in is essential and so is a collaboration with CSR function, HR and all other communities, Leverage AI for Good at organization level so various stakeholders get involved. Employees can become a true force for good and harness it in the right way. In PwC our AI team is working on several initiatives focused on health including wellbeing and mental health, environmental issues, education, gender balance and partnerships for the goals (SDGs 17th).
MD: How would advise companies on assessing the ROI on AI for Good efforts?
MLA: Defining the type of problem and problems to be solved and consider the impact on your business in terms of both tangible and intangible benefits. Have a clear scope will allow defining a solution, its capability, and budgetary requirements. Proper governance is essential for any type of technology application; hence all AIs should use governance across the lifecycle from strategy through execution and use. Have your AI for Good application as part of Responsible AI. Partnership with others to create solutions in an ecosystem approach, bringing together owners of data sets, models, frameworks, computing power, standards, etc. Collaboration & sharing is essential to reduce costs & have a sustainable impact, at scale.
Lastly, standardize and productionalize methodologies to allow for cost efficiency in developing a solution from scratch. There is a lot of business and technology knowledge from the industry awaiting to be implemented in the space. Perhaps the most efficient way of addressing SDGs with AI by 2030 is to foster & create communities for sharing of knowledge, resources & relationships to re-use solutions that work and scale to achieve the right impact.
MD: What are some frameworks or guidelines companies can use for AI for Good?
MLA: Excellent question. IMHO any framework used with success can be adapted to specific AI for Good type of problem. For example, Business Model Canvas tools can be used for social enterprise to define the scope, value proposition & more. When looking at key capabilities required, a growing appetite for sharing, open-source & free usage of those resources
– Data Commons & Data Trusts for data sharing
– NGO special licenses or open access from tech companies
But what is critical is to know how to assemble and use all those tools to create YOUR solutions! Our work at AI Commons focuses on defining a high-level approach to problem-solving with the right tools from adapted frameworks and sharable resources. On top of this, we have a global AI expert community of providing advice and the governance and quality control required.
Founded by Youshua Bengio, Francesca Rossi and Stuart Russell, AI Commons is driven by an amazing bunch of AI big brains and hearts. More details at https://aicommons.com/about-us/
We are currently working on MVP, so stay tuned this autumn to hear how you can get involved.
Lastly, I would like to shout-out to a few other groups doing FANTASTIC work in the AI for Good space, follow them, join and augment their messages! Only together we can truly deliver AI for Good.
- Mila — Quebec AI Institute gathering brightest minds in research to work on the same
- Kriti Sharma leading AI for Good UK
- ITU for leading a truly global revolution on AI for Good