The age of "Assisted Everything"
The age of "Assisted Everything"
Why I predict that by the end of 2024 most knowledge workers will work together with a digital assistant on their daily tasks.
A lot of media coverage in early 2023 around ChatGPT currently focuses on the threat of spreading misinformation or that chatting with it can create creepy results. While these are interesting findings, I have the feeling that they are (for sensationalist purposes?) missing the point of what this technology is actually good at: assisting people in creative processes.
While Bing might currently be mostly interested in the “assisted search” use case, I doubt that this will be the most society and job-transforming use case of ChatGPT in the near future.
What is ChatGPT (and a LLM in general) good at?
In principle, Large Language Models (LLMs) such as ChatGPT use a half-finished text and always predict the next word which has a very high probability of fitting in (great technical write up here). This means that by definition LLMs are great at coming up with creative solutions to how the text should continue.
What this means in essence is that while creative writing is a natural process for this type of AI aspects such as logic, math, truthfulness or politeness don’t come naturally and need to be added as functionalities.
Since ChatGPT is trained on existing data and uses its existing corpus of learning together with some inserted randomness, its results will likely always be fitting roughly into the mainstream idea of what a mediocre text would look like.
What are daily applications for creative writing?
When trying to predict how in the near future the world will likely be disrupted, it makes sense to focus on applications of what the AI is great at already and to assume that this is where the biggest immediate potential lies.
So where do we perform “creative writing” in our day to day activities?
Some cases are very obvious, such as school essays, amateur poetry, wedding speech writing etc. Others are less obvious, such as writing technical product specifications, finding inconsistencies in legal documents, composing music and many more. Let’s dive into some of them and see how the near future is going to quickly be transformed in these areas.
What does Assisted Everything imply?
School and Homework → towards Assisted Teaching and Assisted Learning
Given that is an obligatory task and kids are smart and tech savvy, it is no wonder that ChatGPT technology is used immediately to finish homework assignments in no time. While educators' first impulse was to ban students from using ChatGPT and later some ideas emerged of assignments that cannot be faked, smart teachers and education experts have quickly found that embracing it instead is the right strategy. Teaching students how to use this new technology is essential, since it is here to stay and probably a better idea to teach kids how to use it responsibly than to pretend it does not exist. So teachers are encouraging students to use ChatGPT, but to highlight how they used it in their work.
But the biggest potential for disruption does not lie in how students will game thesystem; it lies in Assisted Teaching, in which teachers will have an assistant at their side helps them prepare better lessons, personalize learning content, make content more contemporary and much more.
Imagine being able to update your high school seniors chemistry class on polymers to include content, references and jokes from the TV show Breaking Bad within minutes.
Imagine that when you are being asked “why do we need to learn about vectors in math class?”, you are able to mass-customize exercises with examples from all your students favorite hobbies.
Imagine being able to teach the differences of Old English dialects using Elon’s tweets instead of 7th century texts.
While inspiration for concrete suggestions for teachers will spread fast, students will also be able to learn more about any topic through Assisted Learning themselves:
Imagine being able to take paragraphs from complex subject matter books and get a simpler explanation in plain language within seconds when you are struggling with a new concept.
Imagine generating for yourself a geography quiz by just writing 5 lines of text instead of going endlessly through flashcards
And if you don't trust the AI to output correct content, teach your students critical thinking by using Assisted Teaching to write history essays and let your students find mistakes in it.
Law → Towards Assisted Litigation
Instead of arguing whether AI is suited to handle legal cases by itself or letting it pass bar exams, the more interesting use case of the current technology is Assisted Litigation. While there are definitively many hard things that lawyers have to do, a lot of work includes summarizing big batches of documents, finding inconsistencies, formulating extensive memos, etc. These all are tasks that require a fair amount of creativity, but allowing a lawyer to edit a “quite good” draft is going to make lawyers far more efficient. So maybe the first use case is actually rather going to be the Assisted Paralegal who relieves the young associates from some of their all-nighter work.
Software Development → Towards Assisted Coding
Software development is probably the most advanced area when it comes to implementation of the Assisted Everything principles:
While sitting together with another human developer to look at the same code and work together (“Pair Programming”) is known for increasing code quality and job satisfaction, it is not widely spread, since it increases the amount of person hours that go into a project.
This is probably the key behind the success of the Assisted Coding tool “Github Co-Pilot” that only one month after its launch already was used by 400.000 programmers. In Assisted Coding, the AI suggests code that fits the context that you are currently writing. It is able to execute tedious repetitive tasks, but also to creatively complete code in languages that the user might not even be very familiar with and propose automatic tests for the existing code.
According Githubs own research, developers that use their Assistant are 55% faster in writing code and feel more productive and fulfilled with their jobs.
Creating Art → Towards Assisted Writing, Composing and Design
2022 already saw a major shift in the visual art world. With the revolution of Midjourney and similar tools first two things happened:
A major debate regarding copyright, ownership, artistic merit and other topics emerged, as users started using existing artists’ names in their prompts and styles were copied
Artists who used these Assisted Painting services regularly actually became very good at combining and remixing, together with the AI and manage to create amazing and unique outputs.
With ChatGPT, a similar shift is now happening in the world of written art. The sheer amount of new services popping up to assist users in writing anything from children's books to condolence cards, from slide presentations to love letters, from witty flirt lines to sermons, is only starting to scratch the surface of what is possible.
Where it gets even more interesting though, is use-cases in which the creative inputs can be text, but the outputs are of another medium:
Drama and Moviescripts are a great example where the written word is later translated into other art forms. Assisted Writing of screenplays today already produces results that make professional writers claim “with a bit of editing I could take that to Netflix: just need to finesse a little bit”.
But even in less obviously connected areas language-generating AI can change everything: Assisted Composing allows audi artists to create music from text.
Even complex visual art forms such as apps or websites might soon be easier to design with an Assisted Designer than by yourself, even if you are a professional.
Hardware Development → Towards Assisted Engineering
Hardware development at first seems like a very math and physics intense discipline that does not immediately come to mind when thinking about LLMs in AI. However, actually a whole lot of steps along the product development lifecycle are text based: Everything from writing specifications and product requirements, to analyzing and reviewing datasheets, preparing test procedures and technical documentation is text-based and can therefore be improved through Assisted Engineering. Since this is a topic that is very close to my heart, I’ll dedicate a few posts to this topic, starting with a general overview of why products that are developed using Assisted Engineering will counterintuitively be safer than todays’ products; even if current LLMs are terrible at math.
Similar to the Assisted writing case, the first applications of LLMs in engineering modify text alone (such as improving requirements, preparing progress reports, summarizing technical datasheets, etc.), but there is even more potential to be unveiled when text describes logical models of our hardware and by altering text we are able to alter the design itself, which I will explore in future posts which you won’t miss if you subscribe to this substack.
Leading Teams → Towards Assisted Management
While it is unclear whether being assisted in writing performance reviews or reviewing applications in Assisted Hiring will make the processes better or meaningless, only time will tell.
However it is very obvious that summarizing calls and drafting minutes of meeting, reviewing KPIs for consistency and clarity, brainstorming pros and cons for decisions, answering questions of newly onboarded staff and similar tasks will speed up daily tasks in management immediately and probably be hard to work without within a short period of time.
Conclusions
Taken AI generated content as-is with today’s technologies are seldom extremely creative but consistently provide above mediocre results if prompted the right way.
So while the outputs by themselves won’t change the world (yet?), the interaction with these mildly creative texts have the potential to speed up human work extremely. They are a great starting point for any process that relies on a loop of rapid creative creation and continuous improvement:
Why should teachers, lawyers, programmers, writers or engineers sit down in front of a blank sheet of paper, if they can work with an assistant who provides them with drafts, improves it according to their wishes and takes over the tedious parts of it?
Ultimately, efficiency in any white collar job depends on the increased rate of iterations. And if we are able to reduce the time of iterations with ourselves (coming up with the first draft of anything) and with others (relying on the Assistants functionalities of reviewing and proposing), we all get to create more in less time.
There is a literal race amongst new startups to provide solutions and tools in every imaginable vertical is what makes me certain that within the coming 2 years most white collar jobs will be assisted.
Personal plug:
Given my enthusiasm for how the future of work will be Assisted, it won’t come to surprise that at Valispace we just released Assisted Engineering functionalities that catapult hardware developers into this exciting near-term future today.