ChatGPT and the most marginalized

The hype about ChatGPT or whatever AI tool is launched ranges from “OMG, the robots are taking over the world” to “OMG, it’s better than sex.” It’s all a bit…hysterical.

And I see very few really informed people thinking about what it could mean for the most marginalized. Given what CLEAR Global does - support people to get information and be heard whatever language they speak - and how we do it - providing a full range of language services and developing smart and innovative language technology solutions using artificial intelligence that is similar to ChatGPT- I think we’re well-placed to reflect on how this new language technology will affect the most marginalized, how it will impact on international aid, and what changes it will make to the way we work.

A bit of background on ChatGPT and similar systems:

• It’s based on predictive text built on VERY, VERY large language models.
• It’s ability to predict accurately has been developed by learning from humans (unfortunately in the case of ChatGPT by paying Africans to work in terrible conditions)
• The robots are not taking over the world
• It’s not perfect - it hallucinates.?
• It doesn’t know what “truth” is. It told me that I was VP of a company (I’m not)
• The biggest risk is thinking that it understands things or has opinions. It doesn’t - it’s just putting together words that it’s been taught.?
• It’s also not evil. It doesn’t have intentions.?
• It can be harmful, though, and it is biased.

This week alone, ChatGPT has probably saved me at least 2 hours of work time.

So, yay!

And that’s what it’s best for. It saves people time. Someone said that people won’t be replaced by the technology; people will be replaced by people using the technology.

But what does it mean for the most marginalized people?

Does it help the most marginalized?

Perfect English isn’t so necessary. ChatGPT and similar technology mean that writing in English and French is a lot easier. The barriers to sounding intelligent and writing in native English or French (or any other language that’s used a lot on the internet) are essentially gone. You can put any text in it and it will turn it into lovely prose - or a poem or technical jargon. It will need proofreading - and you will need to know a good deal about the topic to ensure it isn’t contradicting itself - but written language skills won’t be a problem. This opens access to opportunities that involve writing in English or another major language - funding proposals, jobs, media articles, whatever.

The digital language divide is bigger. While these large language models are great for writing in some languages, language data doesn’t exist for languages that the most vulnerable speak. So they will be left even further behind. If you don’t speak a major language, the barriers to getting information got higher. Because writing in those languages just got easier, those who don’t speak them don’t have access to the tool -? a summary in English of 20 articles in French doesn’t help them. Their ability to get information quickly and compete with everyone else just became immeasurably more difficult.

Digital Literacy is a big deal. In addition, digital literacy - the ability to find, evaluate, and communicate information by utilizing typing or digital media platforms (from wikipedia) - is not as widespread amongst the marginalized. ChatGPT has made people who have not had much access to or ability to engage with digital tools more vulnerable because it is so easy to believe that you’re having a conversation, that it’s a friend, that it listens and gives advice. It does not - and it should come with a warning.

To me, the above are what’s important to be thinking about. And there are some other impacts on the aid world that are worth noting.

How will it impact the aid world?

Grantseeking. Project proposals and report writing are going to get ALOT easier to write. Basically, if it’s been written before, ChatGPT will reproduce it - and possibly better than a human can. Humans will need to review and edit anything ChatGPT does. But, if you know what you want to do and how you want to do it, ChatGPT (or Bard) will write it in perfect English or French.

• And this is good, because so much of getting funding relies on a perfect command of English. ChatGPT could really open doors for smaller organizations that can’t afford proposal writers.

Donors. As a result, donors are going to have to really understand what they need to know, changing their evaluation criteria to focus more on technical knowledge, experience, and an understanding of the context that can’t be sourced from the internet. (Although, it’s important to note that ChatGPT, for example, only has information until about September, 2021. It still believes that the relationship between Russia and Ukraine is “tense”.)

Innovation. ChatGPT isn’t really creative; humans are creative. ChatGPT might unlock new ways of thinking and a different perspective - and that might help a human create. Education for displaced people (those experiencing the myriad forms of poverty) should focus on how to use personal experience to make best use of technology and how it can lead to better, more inclusive ideas.

Feedback mechanisms. Any call centers, hotlines, etc (as long as they are in major languages). It will be much easier to get answers to questions about almost anything, saving time and resources currently expended on overworked feedback workers supporting accountability mechanisms.

Data protection. So many people use the technology without really understanding what happens with the data you input. Basically, it keeps it. This links to digital literacy, of course, but even the most digitally literate - people who write code, for example - have asked ChatGPT to fix their proprietary code. (doh!) It’s easy to imagine a health worker putting a bunch of personal data in and asking for an analysis. The data is then “out there” and can be accessed.

Finally, some thoughts about how this technology changes to the way aid workers could do their jobs.

Learn to use language tech better. Language technology will bring significant changes to the way aid workers do their jobs. For example, aid workers will need to learn how to effectively use language models and other AI-powered tools to communicate and provide assistance to the populations they serve. Additionally, aid workers will need to be trained in data analysis and interpretation, in order to effectively use the insights provided by these tools.

Digital literacy. Language technology will also require aid workers to be more attuned to the potential risks and drawbacks associated with these tools. Aid workers will need to be vigilant about the potential for bias and discrimination, and take steps to ensure that their use of language technology does not perpetuate these issues. Additionally, aid workers will need to be mindful of the potential for job displacement and take steps to ensure that vulnerable populations are not further marginalized as a result of these technological advancements.

Like most people, I think ChatGPT, like other language technology, can be used for good and for evil. And, like most language technology, until the international development sector decides to harness our data appropriately and invest in developing language technology for the most marginalized, the digital divide is just going to get bigger and bigger.

At CLEAR Global, our goal is to work with partners, particularly those based in communities that speak languages that technology doesn’t serve, to create more of this kind of technology for the more marginalized.?

www.clearglobal.org