Specialists in oral translation services

Friday, July 28, 2023

Artificial intelligence in interpreting—and elsewhere

 By Martha Hobart, AIB

Photo by SergeyBitos From Shutterstock

Inspired by Patricia’s interesting post describing her experiments with ChatGTP in last month’s AIB blog, I decided to dive a bit deeper into the fascinating world of artificial intelligence. Where I found a bit of everything, enough to keep anybody online for many hours.

Let’s look first at the current definition of AI.

A basic view taken from various sources:
    1.    The ability of a computer or other machine to perform those activities that are normally thought to require intelligence.
    2.    The branch of computer science concerned with the development of machines having this ability.
    3.    Intelligence exhibited by an artificial (non-natural, man-made) entity.
Plus more.

Some core AI concepts:
Artificial Intelligence vs Machine Learning and Deep Learning:
    •    The term Artificial Intelligence has the broadest scope. It refers to any type of technology that appears to “make its own decisions.”
    •    Machine Learning is one of the methods used in advanced AI solutions. ML algorithms self-improve automatically by analyzing more and more data.
    •    Deep Learning is a type of machine learning that uses artificial neural networks.

If you really want to get into the weeds, try this detailed discussion.

♦️ Resources for AI services & info

Most of this article focuses on AI in general to provide an overview of what it is and what it is capable of. Much of what is written about AI in interpreting (especially conference interpreting) is slanted in such a way as to transmit the idea that it is fully “mature” and says little or nothing about its limitations.

There are dozens of platforms that offer tools & services for businesses, groups and individuals wanting to use AI in a variety of ways.

I stumbled across a platform called Tidio, which offers services to assist e-commerce businesses in communicating with customers using live chat and chatbots.

They set up an experiment involving some 20,000+ internet users in a test of their skills at distinguishing between AI and human creations (photos, artworks, music and texts). The test is still available to anybody who wants to take it and is followed by a detailed report on the overall results of the original participants.

You’ll find it here.

The report on the test results is extremely detailed and informative.

The page has a lengthy list of links that will take you to specific sections of the report, which led me to hours of exploring the net for information on AI itself plus endless ways of using it.

♦️ Some examples of AI

Let’s look at a few ways in which AI can be used, sometimes for serious endeavors and sometimes for other purposes.

Conversation between bots: video
Two AIs Talking To Each Other: Bing Meets ChatGPT
This is a good illustration of the difference between a traditional chatbot (designed to assist users but can also drive them crazy) and a bot created by ChatGPT (designed to engage in human-like conversation).

Conversation between human & bot: video (Distinguishing between human & AI)
Challenges to overcome in creating bots that appear to be human.
I Challenged My AI Clone to Replace Me for 24 Hours 
A video that discusses creating what are known as avatars, designed to make people think they are a real person and not a bot.

So far we’ve seen examples of videos featuring bots. Then there are physical robots — machines designed to look and act like humans. Which is a whole other rabbit hole we won’t go down now.

Producing articles on different subjects: scientific, political, fiction, etc.
It’s not necessary to discuss this here because Pat’s blog post did a thorough job on the subject. Although she focused on ChatGTP, this tool is based on the same underlying algorithmic structure as more sophisticated AI generators.

If you’d like to get a taste, below is a simple online dialogue generator. There are more in Patricia’s article.

Composing music
Here are some examples of music generated by AI at different stages in its development.
2 songs, both from 2016:
Daddy’s car (said to be the first song composed by AI)
Mr. Shadow

From 2021: 
Classical music composed by AI, played by humans

Now we’re into early 2023. Google’s new AI turns text into music, which means you tell the AI what you kind of music you want, and then see what you get. Here are several 30-second audio clips in different styles.
If you want more, just go to YouTube and go nuts.

♦️ The touchy subject of whether AI can be sentient or have emotions.

AI sentience is a hot-button issue that generates a good deal of concern among people who do not have a clear idea of how AI technology works. It is also a fertile field for many media outlets whose favorite pastime is fearmongering.

A recent article on the subject: Is AI Sentient? Could it Ever Be? Experts Weigh In

My opinion is that the real risk in AI—at least for the present—is its misuse by people who use it to manipulate people’s behavior and beliefs. We should remember that tools like ChatGTP function by scooping up information from the internet based on instructions given it by humans.

Sentience is much more than information produced by humans; it requires a clear definition of human consciousness. With this definition in hand, to quote from the above article, "[scientists and researchers] would also need to use that definition of consciousness and translate it into exact algorithms to program into AI systems. Human emotions, perceptions, etc., would have to be deeply understood in a way that makes it possible to give that knowledge to a computer system."

So far, we are a long way from understanding and defining consciousness.

Can emotion be transmitted using AI?

Again, this idea is based on ignorance of how AI works. There are AIs today that can create images of persons apparently showing emotion, but these supposed expressions of emotion are the result of algorithms created by humans that control facial and body expression. In still images it can sometimes be difficult to distinguish between human and AI, but in videos the results can be pretty crude.

♦️ AI in interpretation & translation

And so we come to what most of you have been waiting for: AI in interpretation and translation.

I will not go into the current state of the art because Patricia covered that thoroughly in her blog post of June 2023

At this stage in the history of AI, it’s not possible to draw any kind of meaningful conclusions about the future effects of AI on interpretation. Machine and computer-assisted translation have been around for a long time and could be considered a precursor to translation by AI. I didn’t research this subject at any length. Apologies to translators 😇

Interpreting using AI, on the other hand, is taking off wildly. Let’s take a look at its trajectory.

I started by scouring the internet in general using a variety of search phrases. Most of what I found were articles from as far back as 2014 that talked enthusiastically about future prospects for using AI in both simultaneous and consecutive interpretation. I also found a number of academic studies.

Nonetheless, a number of systems were already being rolled out. Here’s an article from 2019 to give you an idea of where things were then:

And here’s one that discusses how ChatGPT can be used in interpreting, so it must be quite recent (I hate it when they don’t give you the date of publication).

I also did some searching on LinkedIn, using the search phrases "simultaneous interpretation" "artificial intelligence", "simultaneous translation artificial intelligence" and "conference interpreting artificial intelligence" and got a few results. Not much from individual interpreters, some expressing concern about the future of the profession but others saying, more or less, wait and see. There is still a long road ahead for this infant technology, and it’s too soon to feel threatened by it.

A few of the more insightful comments from interpreters.

Laura Heredia Abolafio
- An informative summary of what is happening in the sector after attending a Workshop on Interpreting Technologies.

Cyril Flerov
- In a post describing Google’s monumental flop in the presentation of its experimental AI called Bard, he wonders about similar errors in using this AI for simultaneous interpretation. He comments, "The problem is not in AI making obvious mistakes. The problem is in it saying something that sounds true and plausible but is not true."

Marta Altamira Cabré
- In response to fears expressed by some interpreters of being replaced by AI, she has this (and more) to say: “As I see it, machine interpretation will open up a huge number of new markets but I am also convinced that human simultaneous translation will coexist with it and, I dare say, will continue to grow as it has done with the advent of RSI - contrary to what many believed.”

On the other hand, there is a lot of what I can best describe as hype coming from platforms offering AI interpreting services for SI, who tend to tout their offerings as fully market-ready. I can’t help wondering.

However, in a press release entitled KUDO Unveils KUDO AI; the World’s First Fully Integrated Artificial Intelligence Speech Translator, the company described it as a "pilot version" and stipulated that "…now, in addition to the skills of our professional human interpreters, we have an adjunct AI component in our portfolio capable of real-time speech translation in certain use cases."

They then set out recommendations for when to use AI and when to use human interpreters. Other platforms are not so open.

I couldn’t find any platforms offering a demo of their AI interpreting systems. Again, makes one wonder…

To give a more realistic view of human vs. AI, two top-line interpreters, Barry Slaughter Olsen & Walter Krochmal, did an experiment in which they compared human and AI interpretation, using Kudo’s AI, and discussed the differences they found.

♦️ A few bits & pieces

A group of Redditors devised a prank that tricked an AI-written website and caused a stir among World of Warcraft players.

A teacher discusses his experience with students using AI for writing their papers. Should you try to ban the practice outright? Where do you draw the line on computer assisted research and writing? I am reminded of the time eons ago when calculators first appeared and students naturally wanted to use them in their mathematics classes.

A short film about a man who hires an AI robot to seduce a politician, and what happens next.

I had doubts about including this film because I found it rather rough. But what led me to use it was the kicker that comes at the end. Many times what seems to be a failure of AI is really the result of getting what you didn’t expect.

What the film illustrates is that what we think may be a mistake/failure of AI is actually our lack of skill at formulating our instructions.

As the wise advice goes: Be careful what you wish for. You may not like what you get.

♦️ Conclusions?

I don’t have any wisdom to impart about all the foregoing except that it’s best just to take it easy, learn what you can about artificial intelligence and follow its progress.

And have some fun with it.