AI As My Co-Pilot
Today is Day 18 of EventHub's 2023 Advent Calendar, and I have the privilege of contributing today's content.
This follows the Day 17 contribution by Naoya Yamanashi from EventHub's Customer Success team. He speaks about his experience and impressions of EventHub's startup culture. His experience mirrors my own at EventHub. A culture similar to many of the Silicon Valley startups I have worked for in the past. Key ingredients for success.
A little about myself. I joined EventHub about a year ago to lead EventHub's international expansion team. This is my first year participating in EventHub's Advent Calendar and also my first time posting an article on note.
This article is written in English, so maybe a brief explanation about my background will be helpful. I was raised and educated in the United States and spent my career working for US tech companies, both large and small. I first moved to Japan in 2015 to lead expansion into APJ (Asia Pacific and Japan) for Druva, a US SaaS startup. I decided to continue my career in Japan and came upon an exciting opportunity to join my first Japan-based company, EventHub. Rather than bringing IT companies into Japan, I now have the opportunity to find ways to bring innovative Japan-founded startups into global markets. I look forward to sharing this experience in future note articles.
I am ethnically mixed with an American mother and a Japanese father. Despite my father's native language being Japanese, I grew up speaking English and didn't have a chance to study Japanese until I became an adult. I am working on improving my Japanese, but I am not quite ready to take the challenge of writing this article in Japanese. I'll save that as a goal for next year's advent calendar. However, the theme of this article, AI, will come into play in my Japanese studies.
Overview
This article intends to share my experiences and how I plan to use AI tools professionally and personally. I feel that 2023 is the year that AI made it into the mainstream and since it is expected by many to be one of the most transformative technologies in human history to date, I felt it was essential to understand a little about how capable it is and how it works. I hope to share some of what I learned. This article is written for the average person who is just starting to use AI, probably not someone who is in the field and deeply hands-on with AI.
Important Note:
AI is a very general term, and there are many different forms of AI. This article mainly discusses generative AI and Large Language Models (LLM). For brevity, I may simply refer to this as AI. I mainly discuss and use two tools: OpenAI's ChatGPT and Google's Bard. However, many other companies are bringing capable generative AI engines to market, such as Baidu's Ernie and others. As well as many tools based on these engines (DALL·E, Midjourney, Jasper, and many more.)
2023 - The Year AI Went Mainstream
Machine learning and various forms of AI have been around for years, but it was the public release of ChatGPT and DALL-E by OpenAI in November 2022 that made most of us aware of the actual reality and potential of general-purpose AI. It became the fastest-growing app in history, growing to over 100 million users by January 2023. Like many others, I didn't realize what a game-changer ChatGPT was until I got hands-on with it.
The first time I gave ChatGPT a try was in January of 2023.
I think my first question was, "What is GPT?"
Followed by, "Now explain it to me in layman's terms."
And for fun, "Explain it to me in a rap."
I won't paste the conversation here but feel free to try it out.
Answers to the first two questions can be easily answered with a search engine, but what struck me was how human-like the interaction was. As I engaged in more queries and conversations, the potential was becoming evident.
Looking back through my ChatGPT history, I found the first interaction I used for work. At the time, I was localizing the company website.
Perfect! Copy and paste! Exactly what I needed on the first prompt. This is certainly something anybody could quickly write, but it was able to create quality content in the right tone in seconds, much faster than I could. There are other things I would rather spend time and creative energy on.
AI and Learning (…. Japanese)
As mentioned earlier, I am still learning Japanese, which is proving to be a long journey for an adult human. Although I don't have the bandwidth to study formally, I meet with a teacher 1-on-1 a couple of times a week. We select newspaper articles to read each week and discuss. Recently, I was asked to start writing summaries of the articles. It is a beneficial learning exercise but one I still find very challenging.
ChatGPT and other language model translation tools can certainly be used to translate and do the work for me, but that would take me no closer to my goal. However, what if I were to treat the AI like a secondary teacher? Ask it to check the grammar and wording and explain what I can improve. This allows me to get instant feedback and reinforce things while they are still fresh in my mind. It also lets me use valuable time with my teacher to focus on discussion and other aspects of learning that I can't do alone.
I started using AI a few months ago, and when used correctly, it can be a handy companion for learning. Here is an example of how I have been using it. (Apologies to non-Japanese readers). First, I try to write the summary on my own. I then ask ChatGPT-4 to correct my grammar and explain the reason for each change. It responds by showing the correction and explaining the reason for the change in detail. A partial screenshot from my last assignment.
It feels like I have a language partner by my side 24/7. I just need to formulate the right request in order to get ChatGPT to do what I need.
How Intelligent is AI?
As I started to use AI in 2023, this became a burning question. One of the reasons I decided to write about AI was to use this opportunity to do some research and come to some of my own conclusions. There are a lot of differing opinions in the media about how capable today's AI is, and there is an overwhelming amount of information (or disinformation) to sort through.
One thing I commonly found in my reading was explanations similar to the following:
"At its core, today's AI is essentially a really good auto-complete function. It's trained on massive amounts of text data, and it learns to predict the next word in a sequence based on what has come before."
At it's core, this is what it is. It uses a statistical model to guess what word should come next. However, it definitely seems very "smart" at what it does.
The Turing Test
Rather than sifting through the opinions online, maybe there is a way we can measure AI's capability. The Turing Test, proposed by mathematician Alan Turing in 1950, measures a machine's ability to exhibit intelligent behavior indistinguishable from a human's. The test involves a human evaluator who engages in natural language conversations with both a human and a machine, without knowing which is which. If the evaluator cannot reliably tell the machine from the human, the machine is said to have passed the test.
This should be a definitive test, right? Well, when I went searching I found publications online that firmly said "Yes, ChatGPT passed" and others that firmly said "No". After reading through various arguments, my personal conclusion is that regardless of whether it passed or not, we are certainly getting very close.
The best answer I found was by asking AI itself.
In my own experience, I know that AI can give some illogical and inconsistent answers.
Basic Mistakes
Despite having complex interactions, ChatGPT can make some fundamental mistakes. See the basic word count query below. The correct answer is 24, but it gets it wrong. Interestingly, when instructed to explain the steps it takes, it corrects itself and gets it right. It is almost as if it "thinks" through the problem. ChatGPT also tends to have a confident tone in its responses, which can lead to people having too much trust in its answers. Something to be cautious of. At least it is polite when corrected :)
These basic mistakes are well documented with ChatGPT-3.5, but it looks like OpenAI has been addressing this. When the same query is performed on the latest AI versions, such as ChatGPT-4 or Google Bard, it answers correctly the first time. In fact, they both generate code to solve the problem. Akin to a human using a calculator.
Does AI Understand Concepts
As noted above, today's AI implementations are said to be statistical models that do not understand concepts like humans. If AI does not understand concepts, I figured it should be easy to demonstrate that through some basic queries, similar to the word count issue. I searched Reddit and many other sources, and it was surprisingly difficult to find simple cases to demonstrate it. It took a bit of effort to elicit a response that was not entirely logical. I wanted to find simpler, but this is the best I could find.
The request:
It crafted a fictional tale about Whiskers, the swimming cat (full story attached below). It was a wonderful tale. The cat was endowed with abilities to read and understand language. Creating a fictional story was perfectly appropriate, and a human writer would likely respond in the same way. I then challenged the AI and stated that cats cannot read. It kindly apologized and rewrote the story. However, it proceeded to write the story with the cat still being able to understand human language. I didn't explicitly tell it that cats cannot understand spoken language, but it did not understand the concept of my corrective request in the same way an adult human probably would.
However, I repeated the same experiment in ChatGPT-4, which correctly corrected the story on the first try, the same way an adult human would. It is evolving quickly.
If you are interested, the whimsical tale of Whiskers the swimming cat can be found here. The first and final drafts are included.
How Does AI Work?
Spoiler: I can't really answer that.
Ever since I started playing with generative AI I had a burning question. How does it actually work? It feels so fundamentally different from other AI tools to date.
At its core, generative AI is based on an advanced statistical model, a neural network. This is a science that has been around for decades. What is the breakthrough that suddenly made the latest AI so capable? As I started to research, I quickly realized that the science is far too deep to be comprehensible in a few days of reading. However, I will share what I learned since it may be helpful. Those who work with or know how AI works may want to skip this section.
The following is one of the best explanations I came across. It explains how large language models like ChatGPT work and how they became so capable.
One Word at a Time
At its core, today's AI is a statistical model. It generates responses word by word, selecting the next word based on the probability of what should come next. These probabilities are based on all the content it is trained on. In the simplest terms, this is what it is, but there is a breakthrough that seems to have endowed AI with the more human-like capabilities it has today.
Attention is All You Need - The Transformer
A research paper released in 2017 called "Attention is All You Need" is considered to be the ground-breaking step that resulted in the capabilities that AI has today. It introduced the concept of Transformers. It is the "T" in GPT.
G - Generative: Can create text, poems, code, and other creative formats
P - Pre-Trainined: Learns from a massive set of data
T - Transformer: This is the "Attention." It analyzes the relationships between all words in the sentence to determine the concept of what is being asked and uses this to generate responses.
A simple example of a Transformer:
>> The river has a steep bank.
By comparing the relationship of the words in the sentence, it statistically determines that "bank" refers to a geographical feature rather than a financial institution. It saves this as a context, which it remembers throughout the conversation, influencing the statistical probability of what word to output next.
This is what makes AI seemingly able to understand the context of what is being requested and discussed. This, combined with the massive amount of data used to train the AI, is what yields the convincing results we see today.
Can it Think (Let's get a little philosophical here)
Based on a basic understanding of how it works, we can clearly say that it does not "think" in the same way as humans. It is not conscious, it does not think on its own, and it cannot imagine. However, it does seem to have "grasp concepts and "think". The AI community is divided on it.
The are two camps of opinons:
1. AI simply tricks us into believing it is smarter than it is.
2. If it looks like thought, then it is thought.
In my view? I recognize that AI today is just a statistical model. However, is it really that different from the way humans learn and develop a concept of the world around us?
AI shares a lot of aspects of "intelligence" that humans have today, but it is far from conscious. The AI we use today lacks intent, desire, and emotions. Foundations of what makes us human. However, after understanding Transformers, it seems there are no technical barriers to giving AI some of those traits if someone wants to.
This is why it is important for all of us to have some understanding of what makes AI tick. This will be a technology and topic that defines the next era.
Ok. Let's get onto a lighter subject.
My Co-Pilots
Despite the flaws, AI tools today are incredibly useful if used correctly and for the right purpose. I will go as far as to say that in the history of humankind, there has never been a tool that offered such a potential increase in productivity and access to knowledge. On top of that, anyone with a computer or smartphone can use it for free. It is as if we have a team of assistants, experts, and teachers available to us 24/7. Our co-pilots. It would be foolish not to use them. They are not perfect, but neither are real human companions.
Although I have experienced the potential, I feel that I have not leveraged AI enough in my own life. It is not the first tool I have the instinct to turn to, but my opinion has changed after conducting research for this article. From now on, before I do anything, I will think about how AI can be used to do it better and faster.
Here are a few of the ways I plan to use this co-pilot.
Search and Research:
AI is the new way to search. You can have a conversation with AI to refine and adjust your search to get the information you need faster. You can also ask AI to summarize things into easily understandable content.
While researching this article, I found it much easier to get accurate information when using AI. I've begun to feel that traditional search is getting worse. Search results, more than ever, are filled with irrelevant and sometimes misleading content. I looked into this a bit, and data trends show that the internet is starting to get flooded with AI-generated content at an increasing rate. The phrase "AI killed search" was something I began to see in my reading.
Language Learning - Learning in General
As shared earlier, AI has been beneficial for studying Japanese. It is probably the first use case that got me excited about using AI in my personal life. It is like having a teacher's assistant and language partner available whenever I want. I can ask it to have conversations with me, drill me on specific areas of my choosing, clearly explain things I have questions about, and check my work. It might not be able to correct my pronunciation, but I don't think we are far away from that.
This is also an obvious area where AI can be used for both good and bad. Students can use generative AI to do their homework for them, and it will likely be impossible for teachers to know it. However, if used correctly, AI will be an incredible learning tool. It will have a profound impact on the way the next generation is educated. I'm confident we will figure this out and have a generation that significantly exceeds what we are capable of today.
Content Generation
There is a lot of content I create in my personal and professional life that does not require creativity or original thinking. I can offload much of this to AI, freeing time for more strategic or exciting tasks.
At the same time, I also don't want to become a contributor to non-beneficial AI-generated content on the web.
AI in Event Marketing
AI is described as being the new electricity. It is a general-purpose technology not meant for one purpose but can be used in many applications. Generative AI is already making its way into many products today and will play a role in almost every industry. I see it playing a useful role in my own industry, event marketing. It is already starting to be used for content generation. Still, one can imagine many ways it can be applied to provide better experiences for event organizers and attendees alike as the technology matures.
Pre-Event
Smart content creation: Assist in the drafting of content based on the session presentations, such as landing pages, registration pages, descriptions, and banners.
Personalized invitations: Create personalized invitations to attendees based on their interests and past behaviors.
Intelligent chatbot assistants: Chatbots that can answer attendee questions, guide registration, and provide updates about the event.
During the Event:
Real-time engagement: Track attendee activity in real-time and recommend relevant sessions, speakers, and networking opportunities based on their interests.
Live translation and accessibility: Automated real-time translation for a global audience. Provide transcripts and captions for the hearing impaired.
Interactive experiences: Use AI to moderate and customize Q&A, polls, quizzes, and games to increase engagement and collect meaningful data on attendee behavior.
Enhance security and fraud prevention: Use AI to identify and prevent fraudulent registrations and manage who has access to events.
Post-Event:
Lead scoring and nurturing: Use AI to analyze attendee engagement to identify those most likely to convert and personalize follow-up based on the attendee's interest.
The future of event marketing is exciting. However, as we look at ways to integrate AI into the experience, it will be important that we stay focused on the core principle of adding value for the attendees and "empowering connections that matter."