In 2015 I wrote the article, “Brave New World: The Future of Search” based on the explosion of developments we had seen at the time in conversational search and in personal assistant technology.
The articled explored the evolving landscape of search technology and how it will impact businesses and consumers. While the article provided a solid foundation for understanding the current state of search, there have been several significant software development innovations since its publication that allow us to imagine an even more ambitious future in some respects.
Voice-to-Text and NLP in Commercial Applications
First, the article highlighted the importance of natural language processing (NLP) in search technology, which allows users to interact with search engines using human language. Since the article’s publication, significant advances the quality of voice recognition and transcription, as well as, NLP have made it easier to incorporate into more technologies used by marketing teams to develop websites, mobile applications and other customer engagement platforms such as advanced chatbots and voice assistants. Today, AI-powered chatbots and voice assistants are being used by businesses to deliver personalized and seamless customer experiences, from answering questions to providing recommendations and automating routine tasks.
- “A Complete Guide to Natural Language Processing (NLP)” by DeepLearning.AI: https://www.deeplearning.ai/resources/natural-language-processing/
- “The State of Natural Language Processing: 5 Trends Shaping the Industry” by Dataversity: https://www.dataversity.net/the-state-of-nlp-5-trends-shaping-the-industry/
Deep Learning’s Role in Search Technology
Second, the article discussed the role of machine learning (ML) in search technology, which helps search engines deliver more relevant results by learning from user behavior. Since the article’s publication, there have been significant advances in ML powered by deep learning, a subset of ML that uses neural networks to learn and make decisions. Today, deep learning is being harnessed to develop what are called Large Language Models (LLMs) that show promise in being able to answer a wide range of generalized queries which previously narrow ML models were not able to.
LLMs, including the ubiquitous ChatGPT, have used to deliver more accurate and personalized search results, from product recommendations to image and voice recognition.
- “The Science of Deep Learning” by Richard Baraniuk, David Donoho, and Matan Gavish. PROCEEDINGS OF THE NATIONAL ACADEMY OF SCIENCES, Vol. 117 | No. 48, December 1, 2020.
- “Ethical and Social Risks of Harm from Language Models” by Laura Weidlinger et al. arXiv preprint arXiv:2112.04359, 2021.
Advancements in Predictive Analytics
Third, the article discussed the importance of data analytics in search technology, which helps businesses understand user behavior and improve the customer experience. Since the article’s publication, there have been significant advances in AI-powered predictive analytics, which can help businesses forecast trends, identify opportunities, and optimize experiences for customers. Today, AI-powered predictive analytics is being used in search technology to deliver more relevant and personalized search results, from personalized product recommendations to targeted advertising.
- “What is predictive analytics? Transforming data into future insights” https://www.cio.com/article/228901/what-is-predictive-analytics-transforming-data-into-future-insights.html
- “18 Great Articles About Predictive Analytics” by Vincent Granville https://www.datasciencecentral.com/18-great-articles-about-predictive-analytics/
Chat as an Interface + the Desire for Answers
Clearly there are a couple of things that the article got right: for one, the idea of, less choice and more opinion has been one that has dominated Google’s recent work. This is somewhat seconded by the ubiquity of the chatbot based solution popular in ChatGPT and integration attempts such as those into Bing.
That said, the race to provide authoritative answers is exactly the problem that Google and others who have been working on the technology for a long time seem to have with the technology – it’s simply not ready yet. In a media culture where so many are ready to take what they hear as fact, does this present more of a problem than ever?
Update: This week, at Google.IO 2023, a number of things were announced including SearchLabs which is the aggregation of projects happening at Google to improve Search with AI.
Prompt Engineering: The New Application Development
Prompt engineering has become the defacto term for learning “how to talk” to the computer through structured language to provide it with the best amount of detail to respond to the user query. Through numerous popular examples such as this recently published Prompt Guide from Mark Rollins, one can see that the best “answers” are often achieved when the application/bot in question is given the prompt that includes the ROLE it should act in, the TASK it should perform, the SUBJECT of focus of the query, and the FORMAT that the results should take.
New domain specific software applications are emerging in a wide range of creative, marketing, legal, research and software development fields, where the application becomes the “prompt factory” if you will, and leverages other underlying Large Language Models to provide the answers for them.
Consider a few of the following software applications as very specialized examples:
- GitHub Copilot for AI based pair programming
- Scispace for AI based literature review
- Nanonets for capturing and deciphering information from written documents
For someone such as myself who has grown up learning about many different kinds of application development, I think this is incredibly fascinating but it will also be a big learning (or unlearning curve).
New Frontiers in Multimedia Generation and Search
What may have been most alluring to many of us has been the emergence of new creative generation tools for original images or audio based on prompts, often leveraging existing creative material for inspiration. These models themselves are growing in maturity just as fast if not faster than the chat based models, and are starting to create more social anxiety given their potential to infringe on creative rights and/or to instill fear, paranoia and misinformation.
Also, the development of new image search technologies such as the amazing Segment Anything from Meta’s AI Labs start to imagine a new type of interface where we can outline and visually extract interesting pieces of information or imagery in a graphical way, only to put it on a canvas for use in search or new content creation.
I call these examples out separately, as I think they offer potential to give us entirely new ways to consider user interface design.
So where are we now?
In my view, the last few months have been an incredibly wild ride seeing so many exciting ideas play out in real time. Personally, I am interested in spending more time fleshing out the following areas in continued writing and exploration:
- The unchecked growth of training data and what that means. I wrote about this a bit last week in my Substack Notes, really as a response to some great pieces in Nonzero and Simon Willison’s Blog.
- The dilemma around IP use in content generation and in training data itself. Imagine the impact to the web itself if everyone walled up their gardens in response?
- What does responsible development and use of AI generated content look like?
- What does the future of teamwork look like in organizational functions that will see AI impact most immediately, such as content marketing, legal and other areas?
I look forward to your feedback on this piece and others as we explore it together.
One final thought: It’s a great time to learn
I can remember about fifteen years back when we really saw a renaissance in open source learning around new programming languages and web technologies. It was very exciting coming out of a period of many years where everything required certifications and expensive licensing.
We’re entering another period such as this with LLMs, Machine Learning, and so forth. In addition to being able to sample the latest tooling for free or little money, some of the best universities and professionals are building educational content on these bleeding edge topics and giving it away. This makes it an exciting time for those who decide they want to jump in and harness this tech to accel in their careers.
Thanks and feedback welcome
As with anything in this category, the references become wildly out of date quickly, so please recommend updated reading links and such, and I’ll do my best to keep it up to date. Ironically the authoritative references suggested by ChatGPT on some of the topics I talk about are long taken down and now 404, which I think is hysterically well suited for this situation.