It will be interesting to find through research work whether ChatGPT can provide us with a comprehensive search and supersedes its earlier competitors such as Google Assistant, Alexa, and Siri. It was beyond wild imagination to everyone how the conversational robot was capable of understanding questions and answering them with superb accuracy. The application of ChatGPT in the field of medical sciences is intriguing and promises to open a new arena. As other systems of communication, Chatbot should be applied in medical sciences maintaining ethical concerns and remembering that, at times, ChatGPT can give inaccurate and wrong answers.
It was a remarkable event and a red letter day when it counts on November 30, 2022, when the San Francisco, United States based developer OpenAI (the creator of DALL·E 2 and Whisper) launched this chatbot based on a proprietary license (chat.openai.com) called ChatGPT (chat generative pretrained transformer).[1] In just the initial 4 days, OpenAI has estimated over 1 million users. It has been developed on the fine-tuned approach of transfer learning by both supervised and reinforcement learning techniques, The value of OpenAI has been increased to $29 billion following the introduction of ChatGPT.[2] It is now available as a professional plan for $42/month. The current version is ChatGPT-3, and it has been rumored the next ChatGPT-4 version will soon be available by the end of 2023.[3]
If a person wants to develop an app, it will be required multiple people to hire but ChatGPT is a state of art natural language processing (NLP) by using artificial intelligence (AI).[4] Based on correct prompts, ChatGPT can provide new marketing strategies, and revolutionize websites and blogs.[5] The revolutionary bugs can make a near-accurate functionary in a flash of few seconds. It has taken the world by storm. Even the experienced programmer can be replaced by the ChatGPT. In the reinforcement step, the trainers create a model based on first-ranked responses, which enable the creation of reward models that are again further processed based on proximal policy optimization.[6] These algorithms also negate many unnecessary computational pathways, which had been tested and trained in Azure supercomputer infrastructure by Microsoft.[7]
Limitations
Even if some twisted and wrong questions are being framed, ChatGPT will pick up this wrong code and generate counter argumentative evidence-based replies, which otherwise could have created an error message reply in Python and JAVA.[8] This feature differentiates ChatGPT from its predecessor InstructGPT, which filters harmful and deceitful (including sexist and racist) responses. By upvoting and downvoting, a text field with additional feedback will also be generated. Though it appears that ChatGPT will simulate human conversations, it has multifold and versatile unique features as well.[9] It could have multiple applications ranging from testing questions, writing poetry, playing games, or simulating ATM, and emulating a Linux system. Since the ChatGPT algorithm can prompt based on previous responses, it has a promising role to play as a personalized therapist in the medical field.[10]
The limitations of ChatGPT
Many times ChatGPT gives incorrect responses, nonsense but seems plausible, which in the language model is called “hallucination”
The reward model in ChatGPT can be over-optimized, which may deviate from the purpose. This is based on the proposition of Goodhart’s law.[11]
The program has been designed in such a manner that it has been barred to make any political comments.
The review generated provides longer responses instead of brief and crisp responses.
There could be a potential chance of algorithm bias during training.
The ChatGPT service best works in English rather than in other official languages spoken in the world.
One of the major drawbacks is uneven factual accuracy.
ChatGPT is a popular AI tool that has 100 million users within only 2 months.[12] By January 2023, it has recorded 13 million users. Similar programs such as Jasper have similar facilities like ChatGPT. The founder created this research group with an increase in investment in education and health care and was interested in growth and job creation. This can be viewed as an opportunity. During the recession period, it could have been used as a tool, which can be used for prediction but such a GPT is very addictive. GPT for future business proposals and planning is on its way. The neuroflash has an excellent executive alternative that helps writing skills. The charge of GPTplus will be available soon at a cost of US20/month. The popularity of AI-based GP technology is increasing day by day; even in India, there is a sharp increase in downloads in Google Play. Machine learning has been built upon a technology that enables the generation of the type text following a prompt given by the user followed by excitement with language structures in a flash. Both Google and Microsoft are now working with this new OpenAI-based App, and Sensor Tower is a popular chatbot that has headquarters in Vietnam and was released on December 2022. It has jumped from the 20th position in productivity from February 2 to the 8th position by February 14 on Google Play, as shown in Table 1.[13]
ChatGPT received positive reviews from the leading daily The New York Times, which considered this as the ‘best AI. This was endorsed by another popular newspaper The Guardian to generate ‘human-like text’.[14] Other experts also had given positive reviews about ChatGPT and considered this as smart enough despite its flaws. The co-founder of OpenAI Elon Musk himself has resigned in 2018.[15] Google CEO Mr. Sundar Pichai too had reassigned his teams to explore the counteractive measures to combat this large language model. Many consider that the most popular search engine Google is at threat if ChatGPT takes over and they can not improvise themselves with updated technology. A simulation exam paper was given to ChatGPT to solve in England and Wales, in which ChatGPT scored 42% (below PassMark 55%), which can be considered a reasonable attempt.[16] ChatGPT has the potential to be misused, so utmost digital privacy, vigilance and surveillance, ethical perspectives, and end-to-end encrypted security measures are necessary to combat the potential attacks by malware, software, and pirates in the dark web.[17] Some experts have commented and warned about the potential risks of using ChatGPT, as mentioned in Table 2.
The question-and-answer website Stack Overflow has banned ChatGPT in December 2022 and International Conference on Machine Learning banned the undocumented use of ChatGPT to generate any text in research papers.[18] Mental health-based ChatGPT AI has always been a potential source of approach for mental health professionals. This AI base model of chatbots has been attempted by researchers but criticized for its inadequate potential as well as ethical concerns. It may have the potential to replace therapist and simulate empathy but lack the diagnostic skills and flexibility required by a therapist during human interactions.[19]
ChatGPT was not intended for diagnostic and therapeutic purposes for mental health professionals.
It may be knowledgeable about mental health issues but not foolproof solutions for the ailments. Some researchers have criticized that this is nothing but an AI version of information that is searchable on Google and available in Wikipedia.
ChatGPT, when coupled with other digital mental health tools like woe boat, Wysa etc, can be helpful only if used as a supplementary system with comprehensive mental health care and support systems.
The process of how Chatbot interacts like generating human-like texts and responses has been summarized in Figure 1.
Nowadays, chat has become an important way of communication. With the industrial revolution, chatbots can answer multiple queries with a reasonable success rate in a computerized program. NLP has revolutionized the way of talking and communicating such as Siri Alexa, Cortana, and Google Voice Assistant.[20] The conversation has turned essential changes in human life with a new language technique, which is called AI marker language.[21] Chatbots are mainly of two types: generative-based and retrieval-based. Initially, the tensor flow was introduced in 2015 based on the Google brain team. In an e-mail, the patterns are written, which is followed by using a code to construct a conversation. The practical application intrigued by graph building is where the hidden layers are usually used and then recurrent neural network is being formed.[22] Thereafter, the model selection took place, which is calculated from the flow of characters, and the different new versions of the encoder and decoder models are generated by sequence-to-sequence model. Following this, by the process of bucketing, training is being optimally executed [Figure 2].
ChatGPT has enormous potential to generate NLP in various programming. Thus, the prompt is being created, and open domain question answering and paraphrasing are being done, which is followed by Sentinel analysis.[23] Then, the output is generated either in the table-to-text or text-to-table format. The final steps of processing are machine translation setting, code generation, and code explanation. The program language conversation is important for scrapping unnecessary details and converting HTML to text format. Then, the output styles with the different lists, numbers, and headings are generated in a tabular format output style with various narrative modes, which are formatted in informal and personal modes and finally lead to the customized text manipulation. The basic working principle of ChatGP has been outlined in Figure 3.
ChatGPT is capable of writing phishing e-mails and malware, therefore posing a cybersecurity risk. However, ChatGPT has programming to violate the prompts or commands that violates the law and ethical standards. Even outsourcing of Kenyan workers at a remuneration of $2/h to generate toxic content has been expressed as ‘torture’. [23] To prevent the use of AI in dissertation papers and research work, GPTZero software has been developed by Edward Tian, a high school undergraduate student, which can determine how much text has been generated by ChatGPT. Even by manipulation of prompt engineering (jailbreak), ChatGPT may generate confusing replies on various issues such as how to prepare the Molotov Cocktail bomb, neo-Nazi style of arguments, or the Russian invasion of Ukraine.[24]
ChatGPT has been used to improve the mental health and well-being of people and can be applied to cognitive behavior therapy (CBT). It is effective in randomized control trials for CBT using a natural language rule-based approach, encouraging self-disclosure, self-compassion, and digital counseling.[25]
While opening ChatGPT app after downloading it on an android phone, one can get free and premium access, free trial is continued for 7 days followed by US$45/month for the paid version. The potential advantages of ChatGPT are as follows: (1) 24-h service, (2) quick response, (3) able to answer simple and complex queries, (4) easy communication, (5) complaints are resolved quickly, and (6) easily approachable. The most common uses of Chatbots in the medical field are as follows, (a) medical image analysis, (b) patient and hospital-related data processing, (c) being the virtual assistant to the patient, (d) drug discovery by computational method, and (e) converting paperwork to digital data, (f) pathology and diagnostics, and (g) analyzing disease outbreaks, etc.
Since Turing’s test was advocated in 1950, the developmental trajectory moved ahead with Alice in 1995, Siri in 2010, Alexa in 2015, and finally reaching its present shape by ChatGPT in 2022. ChatGPT by OpenAI already has shown lots of promise and will open a new era of application of AI in the field of mental health. ChatGPT has been used to improve the mental health well-being of people and can be applied for the CBT. It has been found to be effective in randomized control trial for CBT using natural language rule-based approach, encouraging the self-disclosure, self-compassion, and digital counseling.[25]
