Lost in Translation: Linguistics with Mr. Robot
Tomas Haffenden
Creative | Presenter |Producer | LinkedIn Top Tech Voice - Creativity + Technology
Communication has always been our most outstanding achievement as a species. But the way we effectively communicate with each other is littered with linguistic inaccuracies and broken rules. A combination of common sense and context allows us to easily navigate these issues. Still, we are facing a serious challenge to this approach that has the power to undermine the potential of our newest and most exacting technology - Natural Language Interfaces.
What the Hell is an NLI?
Natural language interfaces are not exactly new. Consider the humble search bar a simple example. You type the thing you want to find, and Mr. Robot will go and search for it. Google and search engines were the next iteration of this. This time, you can put in multiple keywords and off it goes with its magic algorithm to find a billion matches and prioritise them for you.
The next evolutionary step on this journey is Generative AI, where you can give highly complex instructions, and Mr. Robot will have a crack at fulfilling your request. The big difference is that we are now in conversation with the technology, meaning we can layer additional requests and build a highly complex set of prompts and an equally impressive set of outputs.
The exciting promise of NLI is that provided you can write a precise enough prompt, Mr Robot can do the rest, meaning it will perform tasks that would typically require a complex and often expensive human skill set to achieve. Non-coders could code, non-designers could design, and as the complexity of integrations grows, so do the possibilities.
This has enormous potential in our ability to create and achieve beyond the confines of our own skill set, provided we can write a clear enough set of prompts, and here lies the challenge. Most of us are not as good at written communication as we think.
Writing Good English Well
Unlike the spoken form, written English is a minefield of complexity. Not only does every word have a precise definition, but it is also littered with punctuation and grammar rules that, if misused, can completely change the meaning of your message.
What's the difference between "Stunning" and "Beautiful," for example? Nope, me neither; I use them interchangeably. However, there is a difference.?
Our generally loose and pragmatic understanding and use of words only become a limitation when we start communicating with someone with superior linguistic knowledge. Historically, it was only a major issue if you were friends with clever clogs like Steven Fry (QI) or Victoria Coren Mitchell (Only Connect) or perhaps you were submitting an English essay. Yet again, some context and a little common sense usually allowed the reader to get the point.??
But we have a new friend devoid of common sense and who only has the context we give it: Mr. Robot.
Chatting with Mr. Robot
As we develop our relationship with this technology, many are starting to report a trusting trend in unexpected and poor-quality outputs, but this points to the linguistic shortcomings of the user more than a failure of the technology.
The cold, hard truth is that we use many words we do not fully appreciate the definitions of. For example, what is the difference between 'Reformat' and 'Refactor'? A lot. 'Reformat' means change only the presentation of something; 'Refactor' refers to doing the same thing differently. These subtle differences are hugely significant to Mr. Robot and his ability to do what you want him to do.
An increasingly common phrase is: "Bad input, bad output."
领英推荐
This is an exciting challenge to the idea that this technology will steal all our jobs and replace us. It certainly has the potential to replace some things we do,?but?only if we give it the correct prompts, which is more challenging than it might have initially seemed.
A Lesson in Precision
A great example of the importance of precision is the eternally amusing "Exact Instruction Challenge" (https://www.youtube.com/watch?v=Ct-lOOUqmyY) used to torture; sorry, I mean to teach children the value of clear instructions.
The owner of the child follows the exact instructions given by the child to do a charming, simple task like making a sandwich. Naturally, some critical 'common sense' instructions like 'take the lid off the peanut butter' are omitted, and we all have a good laugh at how stupid children are.
However, the shoe is now on the other foot, and it's our turn to be the instructor; I dare say Mr. Robot is having a good laugh at our struggles. With no common sense and only the context we provide, we must up our linguistic game.
A whole field has been created, prompt engineering as a profession committed to testing, learning, and educating users on how best to provide the context and instructions to get the most from Mr. Robot.?
Although it is a handy way to jump on board and use the technology without needing to upskill your communication skills, it will not be the long-term solution.
Conclusion
If we are to realise and benefit from NLI truly, the solution is returning to the classroom and filling in those gaps in our linguistic understanding. How? Go and make friends with a prompt-engineer.
However, globally, there is a serious mountain to climb for the 773 million illiterate adults (UNESCO).?
For the next generation, 'Gen AI'(?), Forward-thinking schools will no doubt have rushed to update their English syllabi and would go as far as to predict 'Prompt-Engineering' or 'Robot-Lingustics' (need a better name) will quickly become a subject of study in its own right.
The future will be defined by those who have developed the linguistic understanding to make use of this technology and those who have not. It promises a world where, with a Prompt-Engineer's skill set, you can mitigate any lack in your native skill set and instruct Mr. Robot to do an increasingly complex set of things on your behalf in your new role as CEO of Human Creative Potential Inc.
Communication has always been our most potent human ability and the complexity of how we use it to share ideas is what makes us unique. Still, this new focus on developing our general linguistic understanding beyond its current level will be critical, so resubscribe to Word of The Day and add the complete Oxford dictionary to your Xmas gift list.
Now, go create.
???????????????? ???????????????? ?????????? ????????????... ?? ?????????????????? ???????? ?????????? ?????????? 2018! ???????? ????????, ?????????? ??????????, ???????? ????????. ?????????????????
1 年In my coding days, we used to say with an absolute tone: "Garbage in, garbage out", a deterministic view where any waste or useless material entering a process will "definitively" lead to waste or unusable material emerging. On the plus side, "Bad input, bad output" still allows for some (even though possibly bad) value despite limitations. ??
???????????????? ???????????????? ?????????? ????????????... ?? ?????????????????? ???????? ?????????? ?????????? 2018! ???????? ????????, ?????????? ??????????, ???????? ????????. ?????????????????
1 年I personally think that the onus should not be on humans to fundamentally change themselves simply to interface better with evolving technologies. Rather than expecting people to develop superhuman language skills to compensate for current AI shortcomings, the design philosophy of natural language systems would be better served focusing on helping machines close gaps in understanding nuanced human communication. After all, these interfaces were created to serve human needs, not the other way around. Asking large populations to fundamentally re-adapt longstanding communication styles, contextual norms or spheres of knowledge just to interact with inhuman systems risks disenfranchising certain demographics and compromising meaningful contact. Instead, continued progress in areas like contextual reasoning, commonsense knowledge and nuanced emotional interpretation will allow virtual assistants to progressively meet users where they naturally are linguistically. As AI deepens its flexibility, fluidity and ability to comprehend intention beyond literal syntax, the language divide will shrink. Overall, the aim should be to ensure all people can intuitively access the benefits of emerging technologies through their normal expression.
Software Consultant at 21st Century Software
1 年P'ist (language is defined here, everything else is wrong) vs D'ist (this is today's language, you rules are out of date). Most folks run D'ist, so AI needs to learn to adaptor to be considered a square stink-in-the-mud.
Working to deliver learning outcomes
1 年Perhaps it’s time to revolutionise our education system - although the AI options are available to all it is only those with exceptional linguistic skills that are able to truly utilise its possibilities. Wondering if exceptional linguistic skills are a standard output of the current system?