The Dangers of AI, Both Real and Imagined

AI NLP is a break through for low resource minority languages. Meta and Microsoft are providing open source translation tools for low resource language communities, both of them citing concerns about the threat of language death, among other reasons. Templeton World Charity has targeted 386 minority language with at least one million speakers who don't have access to essential information because of a language barrier. They view AI NLP is a welcome solution for gaining access to essential information.

Thanks for this contribution Michael. I have read it, and followed a few of your suggested links. I also participated in a recent webinar given by missio nexus, and have explored this whole area a little myself.

I am concerned. My 'concern' seems to arise from beyond your radar screen, if I may be so blunt. How to express this bluntly ... The MW (Majority World) these days suffers from many things. One of those is, the tendency for Western experts to tell  MW people how they do things in the West, and then to recommend that they adjust that to their own context. It's always, it seems, the indigenous doing the 'adjusting', which is fair enough in a way - but does this kind of 'adjust for your context' situation actually produce good balanced helpful advice and material?

Imagine a tennis player joins your soccer training match. You give him instructions in how to head a ball, how to dribble, and to aim for the net. Then you say 'adjust this for your own context'. The tennis player goes back to the tennis match. Balls are neither headed, dribbled, or aimed at the net. 'How did it go?' you may ask later? 'Great, your advice was very helpful' is the likely response, from someone used to being respectful to patrons.

Now, chatGPT and such like, has taken things a stage further. Remember - that in its formal circles many MW countries operate using English. Now though, instead of telling them 'adjust this for your purposes', you tell them 'my friend chat GPT will give you the answers in good English prose'. Now - no chance for adaption!!

So instead of the MW person being given some guidelines, and then making sense of them (or rejecting them, or twisting them into a new shape) for their own familiar context, they are required to share them as they are, with no more adaption possible than are found in pre-existing English data-bases. So, the window through which MW people used to retain a little of their own thinking possibility is closed to them, unless they block chat bots, but how can they, and how dare they (as to do so would imply that they are racist, albeit to themselves).

I do not believe that chat bots can ‘do’, say, Tanzanian English. Or Indonesian English, with all its subtle cultural content, arising from how those Englishes are used as translations from their indigenous tongues, and ways of life / cultures, etc. Yet, as I say, if one says AI is only for the First World, surely that won’t be acceptable either … ?! So, from hereon, zero cultural adjustment.

My quick experiment, I asked a chat bot ‘Give instructions to a tennis player on how to head a ball during a tennis match.’ It responded: “I believe there may be a misunderstanding. In tennis, players do not head the ball as they do in sports like soccer or basketball. In tennis, the objective is to hit the ball with a racket, not using any other part of the body.” If tennis is, say, Zimbabwean culture, and Soccer is English culture, is the bot not being racist?

See also: https://www.academia.edu/s/d9eb9fcef8?source=link

Hi Michael,

Apologies if I wrongly implied that your radar screen wasn’t appropriately calibrated!

I think differences remain between approaches you are advocating, and what I am suggesting. I hope you will not be offended in this egg-shell-laced area of thinking.

For example, I do not try to find and elevate examples of MW people’s own prowess at ‘beating the West at its own game’, such as Timnit. In a sense, I don’t think that is fair. It implies (doesn’t it?) that the West is ahead of the game, but others will beat it on its own turf. Is that a reality, or a myth? In my experience, often a myth. I also tend to see this as a bit of an excuse. Pardon me. That is – why do we wait for an Ethiopian to speak out, when ‘we’ (assuming you are a) Westerner(s) should be doing so? In other words: Westerners put their mind to exploring everything they find under the sun … except African people (MW people). They shy away from the latter big-time. That, to me, is an injustice (caused by fear of accusations of racism, etc.), that results in our times at elevating whatever few Africans seem to ‘crack the code’ rather than Westerners getting their own feet wet. So, instead of elevating an Ethiopian who offended google, how does google get away with not having staff in Ethiopia using Ethiopian languages? I don’t mean native born Ethiopians. I mean American-born Westerners. Now there are many reasons they don’t … but those reasons are, to me, terrible, and must be overcome. You shouldn’t go stomping all over a people just because you have the money and the technology.

In my last sentence above, I think both of us will agree. There may be a difference between us in what to do about it. I think – I am saying that the West should be responsible for its actions. That means listening to Ethiopians, yes, … but not those who speak English who have already become super wealthy and famous by being employed at google. The latter is, is it not, the West only every wanting to hear its own language? ‘We’ should be listening to Ethiopians in their own language(s). And, that is not to be in a zoom meeting or a think tank in Washington, or through AI translation. Languages, and the people who use them, are more complex than that. To ‘know’ Amharic, is to learn Amharic as Amharic speakers use it. That is a world apart from learning Amharic as if it is a translation from formal English.

So yes, AI promises to add more levels of imperialistic domination. But I question – will extending the blanket of AI over indigenous languages really help? (That I believe is a project SIL is engaged in.) In other words: I am questioning, often, the desirability and possibility of translation. You know of course that languages are an engagement with a context(s). Hence my example of soccer and tennis above. Because languages do not translate into one another, we shouldn’t try to force them to do so. (The one amazing exception, thanks to the work of the Holy Spirit, of course, to this rule, is the Bible.)

(If ‘harming’ the MW is paying them only $2 per hour, … but that’s a good rate of pay in Kenya … context again.)

You may find this link interesting and helpful as you consider why this technology might be open source at this time. As stated in the article, in the old days there was a lot of conflict between those promoting closed source proprietary solutions and those promoting open sourced (let everyone see what were doing) solutions.

Then companies found that in the early stage of technology development, it is very helpful to be open sourced. It allows people to contribute who are not on the payroll and it allows the solutions to be improved when we don't really know what the best solution is to a particular problem. In a number of cases, insights gained from open sourced solutions eventually are included in proprietary solutions.

One question is always sustainability. There are real costs in providing large language model solutions. The company is willing to front these costs because they believe the solution will eventually be worth having. One motive can of course be to create good will for their brand but often it is more than that in that they believe there will be an economic benefit of having the solution for their company.

So if you look at the history of all open sourced projects, some of the most successful ones have had a company behind them providing resources both in the form of servers and sites and also in terms of people and governance.

Most of us who have seen a number of technology changes realize that technologies can be used for good or evil and that technological change will change societies, communities and minority groups. Often if the groups are asked, they are actually in favor of these changes. It is likely true that they don't fully comprehend the changes the technology will bring (do any of us?) but it is also not right for any of us to think that we can or should override the choices people are making.

Personally, I believe God has worked through technology change to accomplish his will. (Perhaps the printing press would be a good example?) I believe that to the best of our ability, we should use whatever technology change that happens to benefit people and the kingdom. I believe God allows (and may bring about) technology change so that it can be used to benefit his kingdom work. Other nefarious uses of that technology proves again that anything can be corrupted by sin and our sin nature.

Hi All,

Thanks for feedback. Greg – I am not surprised by the feedback that you are giving me. It is widespread.

One thing I thought of … let’s imagine You called, or you found yourself living next door to a couple. They come from another country, so speak another language. The wife does not know English, but the husband does. The husband often chats with you in the driveway. He tells you what they need, please give us some bread, etc. etc. At night you are alarmed – that the wife seems to cry a lot. The husband explains – she gets hungry. You give him more food. She still cries. So it goes on.

Now, one might ask, how long should this situation go on, before you decide to try to do something about it? I mean – you can’t communicate with the wife, and the husband says she’s OK but needs food, then you give food, but every night you hear her crying? I think there will come a point where you will encourage your wife to make a lot of effort (as needed) to learn the language of your neighbours, so that one day, gradually, she will be able to get towhere she can talk to your neighbour’s wife, and find out what might be wrong …

The above illustrates what I seem to be talking about.

I am not saying ‘Africans shouldn’t use AI’, but I am suggesting, and I am not alone, that if they use it, it may in the long term not be at all helpful to them. I concede that this leaves us in a jam. That is not a good jam. It is, in a sense, an outcome of the reluctance, or even refusal, of Western people to accept that African people in a sense are even human … until they learn English and step onto our soil. I mean – when God came amongst mankind he didn’t confine himself to talking to Kings and ignore everyone else. He cared for everyone. And he said, even, ‘let the little children come to me’. Even children! So then, what good reason do we have not to take an interest in African and Majority World ways of life?

You say we shouldn’t be ‘excluding the voices’. This might imply that we should not engage in English, as not-educated Africans don’t know English? Sorry – but you don’t kill the goose – just because they don’t know English is  not a cause for us not to communicate amongst ourselves. What should we be communicating amongst ourselves? ‘How to love the people of Africa’. NOT – (primarily) how not to upset those indigenous brokers who stand between us and the people.

So, it is immoral, I am suggesting, indeed, to exclude African people from our discussion. I.e., they also need to be engaging in this discussion – in their languages, in ways they understand, not only by saying ‘yes’ to us. That requires Westerners to become familiar with those languages and ways of life, so that they can see how AI will or will not impact on it. At the moment – we really do not seem to have any such people. Why?

I hope that makes sense.

Jim, Perhaps others understand your reply better than I do but it seems to be based on a misunderstanding of AI and the Large Language Model technology.

The technology is not only using English. It is not only using majority languages. Facebook has trained a model using the technology on all the translation work we have completed for which audio and text exists. They claim it can process 1100 languages and recognized 4000 languages.

The technology is developing at a rapid pace. I more liken it to the introduction of the Internet. When it was introduced, Christians decried the dangers. Many of these dangers have been realized. Yet for all the dangers, there have been many positive benefits. In fact, my guess is most of us depend on it for our work. This thread depends on it!

AI is much broader than the large language models, but the large language models are the latest and most impactful application of AI.

It isn't there yet, but it may happen soon that your neighbors wife will be able to talk into a device and it will translate what she says and give it to you in a language you can understand. By using such a tool, you can dialog and clarify elements of the communication that may be misunderstood and much more quickly understand the core issues. These are the kinds of changes we are witnessing.

The technology has the potential to bring a lot of benefits to minority peoples. And yes, there are risks that this technology will be used in ways that are harmful to them too.

There will likely be some regulation. One of the curiosities of what we have witnessed is that the big tech players were researching and pursuing the technology but the breakthroughs came from much smaller players and the big tech companies are now asking for regulation. I half wonder if they aren't trying to slow others down so they can catch up ... but that is total speculation.

It is unlikely that anyone will stop this technology from impacting minority peoples any more than you can stop the Internet from impacting them. We can't even stop it from impacting us and our work. It will likely have huge impacts on both.

It will probably never be capable of doing Bible translation per se but it will be used in the process of translation to speed the work and improve quality. It may be able to help make commentaries and other Scripture study and devotional materials available to people who have had very little before because it was cost prohibitive to make this content available. These are just a couple ideas not an exhaustive list. 

So let us live by faith. The wheat will grow with the tares. We need to let God's angels separate the good from the bad at the end. For now let us do good and love all people.