Edited by Lizzy Yee.
In the early days of computing, machines were primarily used for carrying out daily tasks like calculations, while more complex ones were equipped to break enemy ciphers during WWII. The advancement of technology was proven by the increasing usage of computers within a century. In 1966, the term “chatbot” was introduced by Joseph Weizenbaum when he invented ELIZA, which functioned as a psychotherapist. Since then, Chatbot Technology has undergone significant changes because of the emergence of various machine learning algorithms. The invention of Apple Siri in 2010 caused dissenting voices among the public, debating its accuracy and concerns over personal safety issues. In November 2022, the launching of ChatGPT garnered global attention, breaking the record for the fastest-growing user base.
In this age of technology, the rapid development of Artificial Intelligence (AI), particularly in the Natural Language Processing (NLP) field, has played a crucial role in various industries. Based on a study conducted by Juniper Research, the use of chatbots in the banking sector has successfully cut operational costs by up to 7.3 billion dollars globally by 2023. They were trained with massive amounts of data like credit scores, transaction histories, and financial news. The chatbots are able to analyse the information received and produce relevant (accuracy of responses are not guaranteed) answers to every client’s inquiry. The press release also detailed the use of AI in managing insurance claims, leading to increasing customer loyalty. The NLP system also plays an essential role in the healthcare sector. One of the more significant utilizations is to manage the running of the International Classification of Diseases, Tenth Revision, and Clinical Modification, known as ICD-10-CM. It is a system used to classify and store the diagnosis code to ease the physicians’ treatment in similar cases in the future. Zooming in on our daily lives, voice-search engines like Apple Siri and Google Voice utilise the NLP system to shorten the Time to First Byte (TTFB) to 0.54 seconds compared to the previous interval of 2.10 seconds, based on research carried out by Brian Dean from Backlinko. These examples show how humans employ machine learning, a main branch of AI and the backbone of ChatGPT.
Figure 1. Transformer Architecture. Retrieved from https://medium.com/@zaiinn440/attention-is-all-you-need-the-core-idea-of-the-transformer-bbfa9a749937
ChatGPT was built from transformer architecture, a deep learning model that was first introduced by Google Brain via their publication “Attention is All You Need” in 2017. Transformer neural networks use self-attention mechanisms to encode and decode to produce outputs with high accuracy. Let us use a simple example to illustrate how transformers work. The sentence “Where are you from?” will be broken up into "where", "are", "you", and "from". Unlike supervised machine learning, the self-attention mechanism enables the four separate words to communicate. Here comes the technical part. Vectors representing query, key, and value are generated, followed by multiplying the query and key vectors to produce a dot product, also known as the attention score. The attention score is then scaled down by dividing the square root of the dimensions of keys, which then undergoes normalisation using the SoftMax function. The equation of the SoftMax function is shown below.
Figure 2. Equation of the SoftMax function. Retrieved from https://www.turing.com/kb/softmax-multiclass-neural-networks
Now, the sum of the score is equal to 1 which means that the higher the score, the higher weightage of the word. Then, the dot product of the value vector is calculated with the output of the SoftMax function to produce z, the output of the attention mechanism.
Figure 3. Output of the attention mechanism. Retrieved from https://medium.com/@zaiinn440/attention-is-all-you-need-the-core-idea-of-the-transformer-bbfa9a749937
Due to their ability to compile an enormous amount of information within seconds, chatbots have become so powerful that there is a possibility for negative ramifications. Furman University, Assistant Philosophy Professor; Darren Hick, caught one of his students using ChatGPT to write a 500-word essay on the 18th-century philosopher David Hume and the paradox of horror. It demonstrates the ability of the chatbot to perform academic tasks which may be misused for cheating purposes. Recently, the sentence “I want to destroy whatever I want,” produced by Sydney, Bing’s AI chatbot, went viral on the Internet. It illustrates the underlying virtual threat of a non-living, non-matter object to us. This abnormal answer alarms us to highlight the importance of recognising and filtering out the potential risks to improve the results.
In short, is ChatGPT our new friend or hidden enemy? No one knows. It will be a new norm to reside in a virtual global village whilst still living in the physical real-world. The strength shown by ChatGPT will not obliterate the traditional jobs, leading to permanent unemployment among the workers. Instead, new jobs will pop up, akin to how computer science and data-related positions appeared in the 21st century. In the news release after the 2020 World Economic Forum, it is estimated that approximately 85 million jobs will be displaced over the next five years, but on the other hand, about 97 million positions will arise in the robot revolution.
References:
Jonathan, Y. ‘I want to destroy whatever I want’: Bing’s AI chatbot unsettles US reporter (2023, Feb 17). The Guardian. https://www.theguardian.com/technology/2023/feb/17/i-want-to-destroy-whatever-i-want-bings-ai-chatbot-unsettles-us-reporter
Juniper Research. Bank Cost Savings via Chatbots to Reach $7.3 Billion by 2023, as Automated Customer Experience Evolv. (2019, Feb 20). Juniper Research. https://www.juniperresearch.com/press/bank-cost-savings-via-chatbots-reach-7-3bn-2023
Amanda, R. Recession and Automation Changes Our Future of Work, But There are Jobs Coming, Report Says (2020, Oct 20). World Economic Forum. https://www.weforum.org/press/2020/10/recession-and-automation-changes-our-future-of-work-but-there-are-jobs-coming-report-says-52c5162fce/
Brian, D. Here’s What We Learned About Voice Search SEO (2018, Feb 28). Backlinko. https://backlinko.com/voice-search-seo-study
Alex, M. Professor catches student cheating with ChatGPT: ’I feel abject terror’ (2022, Dec 26). New Your Post. https://nypost.com/2022/12/26/students-using-chatgpt-to-cheat-professor-warns/
Ina. The History Of Chatbots- From ELIZA to ChatGPT. (2022 March 15). ONLIM. https://onlim.com/en/the-history-of-chatbots/#:~:text=ELIZA%20was%20the%20very%20first,that%20it%20mimics%20human%20conversation.
Stefania, C. The Transformer Attention Mechanism. (2023, Jan 6) Machine Learning Mastery. https://machinelearningmastery.com/the-transformer-attention-mechanism
Sindhu, S. If you still aren’t sure what ChatGPT is, this is your guide to the viral chatbot that everyone is talking about. (2023, Mar 1). Insider. https://www.businessinsider.com/everything-you-need-to-know-about-chat-gpt-2023-1
Giuliano, G. How Transformers Work. (2019, Mar 11) Towards Data Science. https://towardsdatascience.com/transformers-141e32e69591
ICD-10-CM Codes Lookup https://www.aapc.com/codes/icd-10-codes-range/#:~:text=The%20International%20Classification%20of%20Diseases,external%20causes%20of%20injuries%20and
Medium. (n.d.). Attention Is All You Need: The Core Idea of the Transformer. https://medium.com/@zaiinn440/attention-is-all-you-need-the-core-idea-of-the-transformer-bbfa9a749937
Turing Enterprises Inc. (2022, May 28). How to use Softmax function for multiclass classification. How to Use Softmax Function for Multiclass Classification. https://www.turing.com/kb/softmax-multiclass-neural-networks