The  first in-person Technology Salon DC in three years was convened on January 25, 2023, to pose the question: Can Generative Artificial Intelligence Technology Improve Aid Outcomes? Thought leaders and decision makers across the international development space shared their optimism, skepticism, and uncertainty regarding generative AI and its consequences as we move closer to the uncanny valley.

The moderated and free-flowing discussion was informed by four experts in artificial intelligence uses for humanitarian aid:

What is Generative AI and Why Does It Matter?

Generative artificial intelligence uses AI and machine learning algorithms in order to generate new content such as text, images, audio, video, simulations, and code. ChatGPT—short for Generative Pre-trained Transformer—is one of the most well-known examples of generative AI, and, since its public release by OpenAI in late 2022, it has been used by millions of people to create blog posts, write literary essays, and even create pop songs.

Generative AI such as ChatGPT and DALL-E 2 have been both lauded and loathed across media outlets as more users and uses emerge. Generative AI is already changing the nature of creative work, increasing the accessibility of academic research, starting to anger publication authors, and worrying white-collar workers.  That’s even before AI researchers outline problems of mis- and disinformation, bias, and the discouragement of creativity.

Participants in the Technology Salon recognized similar benefits and risks to Generative AI, and they discussed its use in the development context, specifically the good, the bad, and what it means for the future. (Read our Salon Resources Document for more links)

Positive Generative AI Uses in Humanitarian Aid

The present state of health, education, agriculture, and overall social and economic advancement in low and middle income countries is a problem. In the humanitarian aid sector, we have a duty of care not to make it worse – and to do whatever we can to make it better, now. Digital development practitioners often ask the question: How can we use new technology tools to make life better for our constituents?

Generative AI can be one of the tools we use to improve people’s lives. Tech Salon participants shared many anecdotes that made us hopeful about the potential of generative AI in international development.

For example, singers can use Auto-Tune for better song performance. Although some argued that a singer who uses Auto-Tune has lower vocal abilities than a singer who doesn’t, using Auto-Tune AI benefits artists who may not have the same opportunities otherwise. A performer who cannot sing as well as Whitney Houston but who can entertain a crowd like the Beatles can now sell out concerts with the help of AI.

This change in skillset can be true for international development workers and their constituents too. For example, AI can have an exponential impact on report writing by improving their creation and constitution. We can spend less time on writing and re-writing documents and more time on technical implementation. Constituents with little to no experience writing with proper English grammar can use generative AI as a learning tool and guidance to increase their writing confidence.

In fact, Generative AI is already transforming translation capabilities. Today, tools like Google Translate can instantly identify and translate between languages. Websites, documents, even verbal speech can be translated in nearly in real-time, allowing us to deliver critical information to crisis-affected people globally. One application of this technology is Wysa, a conversational AI for mental healthcare that guides users through therapeutic alliances that are clinically proven to be equivalent to a human therapist within the first week.

Other use cases and upsides shared during the Salon included environmental modeling and planning, and disaster response and reconstruction. One use case that has interesting potential is using Generative AI to predict future actions or repercussions. If Generative AI can be trained to predict what text, audio, or visual humans want to be created next, it could be trained to predict human actions too.

For example, what impact will legislation have on constituencies, and how should they lobby their leaders? Or where will corruption happen in government contracting? AI is already predicting health issues and treatments.

Generative AI Downsides in International Development

Like any new solution, there is a learning curve to understand how, when, and where to apply the tool. Generative AI is no different, and Tech Salon participants noted concerns with it, especially when anecdotes like those shared earlier are not necessarily indicative of reality.

Returning to the example of the Auto-Tuned singer, a change in skillset implies that there is a change in value placed on skills. Some become more valuable, and others become less. This can devalue skills and raise questions like: Is a singer considered a singer if they are not a vocalist? Should they instead be referred to as entertainers? At what point is a skill undervalued?

Many children today rely on digital clocks. As a result, they cannot read an analog clock or understand what direction “clockwise” is in the physical world. Relative to their parents, who can read analog clocks, these children possess one less skill, but is this skill a tragic loss? Do they need to know about fax machines, or we about punch cards? Perhaps not, but it does illustrate how AI technologies can become a crutch. Generative AI can write an essay that is used to inspire new thought, as a reference to improve English grammar, or simply submitted as an original, final work.

The work generated is often biased from the start too. Both data accuracy and representation in data has long been a struggle with new technologies. Generative AI based on biased training data produces biased results. For example, Generative AI translation abilities may be great in English, and produce inaccurate Hindi or Xhosa results.

Sadly, many Generative AI creators are based in Silicon Vally, which is not representative of the United States, much less the world at large. This results in developing country issues, if recognized at all, having little to no consideration or influence in solution development and Generative AI not serving the needs of those in the Global South.

Additionally, there is great potential for malign actors to use Generative AI to create mis- and dis-information. Indeed, many political groups are already utilizing new AI tools to negatively influence perceptions, vaccinations, and elections. Worse, researchers have found that Generative AI videos – Deep Fakes – deeply fake people in developing countries.

Is Generative AI Good or Bad for Aid Programs?

Generative AI will only get better and more ubiquitous. The good and bad potential cannot be ignored. How can we as international development practitioners ensure that Generative AI creators build trust and responsible practices into their tools?

First, we have a moral responsibility to engage with and include local perspectives and bring these outlooks to Generative AI leaders. We have lived experiences that validate the need for local voices in technology solutions. We can also drive decentralization and support local actors to ensure accuracy, representation, and relevancy in Generative AI.

Next, we should build an understanding around AI technologies across our organizations. We may understand AI, but the international development community is often a technology laggard. We can utilize existing tools like the Digital Principles to improve how AI products are understood and incorporated into our work.

Finally, we should adopt AI as a colleague rather than a challenger. AI can be a tool to help both us and our constituents act efficiently, think creatively, and ultimately, create a more prosperous world.  We’ve accepted Internet access, the mobile phone, and even the written word as relevant solutions to complex and dynamic needs in advancing social and economic development. We can now accept Generative AI too.

Written by Rali Sloan for the Technology Salon

Leave a Reply

Your email address will not be published.