Introduction

Simply put, the progress of artificial intelligence (AI) over the past few years has been so good, it begs the question, what changed?

2011 - 2016

Between 2011-2016, we saw the launch of several voice assistants from Apple’s Siri, Amazon’s Alexa, and the Google Assistant. These represented huge investments for these big companies, that they hoped to capitalize on through selling future products and advertisements, and thus bet big with investing into this tech. A key to making a good voice assistant is making it understand speech to capture the initial request, and so a lot of progress was made in voice-to-text (VTT) and natural language processing (NLP), allowing devices to accurately capture speech, translate it into a properly formatted request; then a large database of responses is queried to act on the request. However impressive these voice assistants appeared at first, users quickly found they could be stumped quite easily, often causing the assistant simply run a web search for the users’ question. This sort of failure takes us out of the illusion of talking to an intelligent computer and reveals the lack of depth these assistants provide.

2018 - 2020

That voice assistants are often simultaneously impressive and disappointing is an apt analogy for this era of applications claiming artificial intelligence. Even in 2018, large language models (LLMs) were still in their early stages of development. OpenAI’s GPT-2 was a significant milestone at that time, showcasing the potential for AI to generate human-like text. In their infancy, LLMs faced many challenges such as bias mitigation and improving general understanding of diverse topics. They also required a good deal of pampering, asking questions just right to get the desired output; in this they were quite similar to their voice assistant predecessors. Where they differed, I think, is in their ambitions. OpenAI and their competitors in AI were not resigned to simply build the biggest question and answer database. They sought to create a tool that could actually formulate its own answer given enough training. In much the same way that an experienced worker can answer questions without referring back to the user manual, they sought to make a machine smart. The challenges in doing so are many, and we discussed some of these in our past article From Sci-Fi to Reality: A Brief History of AI .

2021 - 2023

Let’s talk about how we made that last big push to bring us where we are now. A variety of technological, scientific, and commercial advancements culminated in a huge leap forward for AI.

  1. Increased computing power: The rapid development of GPUs and improved processing capabilities have allowed for faster training times, leading to more efficient model execution.
  2. Availability of large datasets: Access to vast amounts of text data from the internet has enabled AI models to learn complex patterns and generate diverse content.
  3. Advances in machine learning algorithms: Researchers have developed new techniques such as transformer architectures that significantly improve the performance of natural language processing tasks.
  4. Open-source projects and collaborative research efforts: Shared resources, libraries, and tools like TensorFlow, PyTorch, and LLAMA have accelerated AI development by facilitating collaboration among researchers around the world.
  5. Commercial interest in AI applications: The growing demand for AI technologies in various industries has driven investment in research and development, leading to rapid progress in the field.
  6. Lack of legal & regulatory interference: As a quickly growing sector, it often takes a while for laws and regulations to catch-up. 2023 has brough with it a number of lawsuits and I’m sure there are more to come, but the general lack of oversight here contributes to rapid growth as well.

Conclusion and looking ahead

We find ourselves today just past the precipice of these huge gains in AI development. The venture capital money is flowing, mergers & acquisitions are commonplace, and the sky seems to be the limit. And yet, much of society is unaware that these huge developments just happened. For examples of this, schools around me have not created AI policies in their handbooks yet. Most of the friends and family I’ve talked to have not installed ChatGPT on their phones (yet), or tried any other AI tools. My employer seems to be representative of most large companies. It has asked that employees not put company data into public AI tools, and is working on some internal AI projects. AI tools have been blocked by the corporate firewall. There is no clear indication of how or when employees will be permitted to use AI in support of performing their roles, something I expect nearly all companies will eventually allow in much the same way they allow the use of search engines. To wrap-up, let’s pull out a crystal ball and make some educated guesses about what will come next.

  1. AI will continue to grow and be included in more and more products. From web browsers to office programs to programming IDEs, companies will continue to make AI more and more readily available.
  2. AI will get more expensive. We are probably still a few years out but remember all those venture capital investors? At some point everyone will be asked to pay up, and that will mean more subscriptions, paid services, or personalized advertising.
  3. Workers and students will try to improve their work life balance, while companies and schools will try to maximize their output. The end result should be a compromise with improvement for both.
  4. Office roles will dramatically change. AI will provide a virtual assistant for everyone. Managing calendars, capturing meeting notes, providing summaries, and drafting emails. Even more dramatic is a software developer role where instead of writing code you are reviewing code generated by AI.
  5. Teaching and learning will dramatically change. Homework as it stands today will be largely unsustainable. Questions that can easily be answered by AI will provide little value. Rather, students should leverage AI. They could be asked to review an AI generated book reported for errors. They should ask AI to tutor them in subjects where they struggle. Do more homework in class, and more learning at home. Ultimately, society will change. Those who try to rope off AI will likely face disappointment, and those who embrace it will need to iterate from lessons learned.

If you find this site helpful and would like to help keep the lights on, please consider ☕buying me a coffee, thanks!