Strong & weak AI
Artificial intelligence (AI) may be divided into two broad categories. Strong AI, also known as artificial general intelligence, refers to the ability of machines to display general intelligence, of the kind displayed by humans, which can be applied to many different tasks and situations. While research is ongoing, strong AI remains elusive, and for now is mostly the stuff of utopian or, perhaps more often, dystopian science fiction. It has recently been suggested that the development of strong AI cannot occur through current large language model (LLM)-based generative AI alone, but may require embodied AI like robots that can interact with and learn from the real world.
Weak AI, by contrast, refers to the ability of machines to apply intelligence to one specific task type, typically involving very specific pattern-matching; such AI often exceeds the ability of humans in a given narrow domain. Examples of weak AI already in everyday use include image recognition, speech recognition, natural language processing, automated translation and learning analytics. Considerable progress has been made in weak AI since the advent of machine learning, and especially deep learning based on artificial neural networks, resulting notably in today’s large language models (LLMs) (see below).
Conversational & generative AI
Within the domain of weak AI, it is important to distinguish between conversational AI which is trained on large datasets of human interactions and provides responses in a limited series of conversational turns (e.g., first generation digital assistants like Apple’s Siri and Amazon’s Alexa, and many corporate or organisational chatbots), and generative AI which is based on LLMs, trained on huge datasets of language use, and generates new (or at least remixed) content in response to user queries; these LLMs include:
-
- OpenAI’s GPT
- Anthropic’s Claude
- Apple’s MM1 & OpenELM (on-device LLM)
- Google’s LaMDA and its successor, Gemini
- Inflection AI’s Inflection
- Meta’s Llama
- Mistral’s Mixtral [France]
- Alibaba’s Qwen [China]
- Baidu’s Ernie [China]
- SenseTime’s SenseNova [China]
- Tencent’s Hunyuan [China]
The newest wave of AI involves agents which combine the ethos of conversational AI (in the form of chat interfaces) with generative AI; these generative AI chatbots include:
-
- OpenAI’s ChatGPT
- Anthropic’s Claude [the same name as the underlying LLM]
- Google’s Gemini [also the same name as the underlying LLM], formerly called Bard
- Inflection’s Pi
- Meta’s Meta AI Assistant
- Microsoft’s Copilot [underpinned by GPT-4 Turbo], which runs alsongside its Bing search chatbot
- Perplexity AI
- Alibaba’s Qwen-Chat [China]
- Baidu’s Ernie Bot [China]
- SenseTime’s SenseChat [China]
- Tencent’s Hunyuan Aide [China]
For links to these and similar services, see the section on Generative AI beyond ChatGPT below. The distinction between conversational and generative AI is widely discussed in the technology press; for examples, see:
-
- A Complete Guide: Conversational AI vs. Generative AI (Matthew McMullen/Data Science Central, 2023)
- Conversational AI vs. Generative AI: What’s the Difference? (Amanda Hetler/TechTarget, 2023)
- Differences between Conversational AI and Generative AI (Sridhar CS/Purple Slate, 2023)
- What is the Difference between Generative AI and Conversational AI? (Webio, n. d.)
For a comparison of the performance of different AI models/LLMs, see the Artificial Analysis website.
ChatGPT and the rise of generative AI
As noted above, the term generative AI refers to a new generation of AI that can generate text, audio, images, videos or other media artefacts. Such generative AI is often combined with a conversational AI interface, with the first well-known example being OpenAI‘s ChatGPT, which essentially grafted a chatbot onto an existing LLM, namely GPT-3.5, and captured wide public attention on its release in late 2022. More advanced versions have subsequently been released, including GPT-4 in March 2023 and GPT-4 Turbo in November 2023; the latter now underpins the paid version of ChatGPT, known as GPT Plus. Work is reportedly underway on GPT-5.
While ChatGPT did not originally have live web access, it now does. It is also now available in app format for both iOS and Android devices. In November 2023, OpenAI made it possible for users with no programming knowledge to build customised GPTs; for a list of GPTs that have been already been created, see The Rundown’s GPTs directory, or go directly to the GPT Store launched in January 2024, or check out the list of Copilot GPTs on the top right of the Copilot home screen.
GPT-4 has been incorporated into a number of other platforms and apps. Importantly, Microsoft’s Copilot is based on GPT-4 Turbo. GPT-4 also powers a Socratic AI tutor in Khanmigo (from the Khan Academy educational service) and in Duolingo Max (from the Duolingo language learning app).
Chatbots > Generative AI chatbots with similarities to ChatGPT include Anthropic’s Claude, Google’s Gemini, Inflection’s Pi, Meta’s Meta AI (which is accessible in some countries via Facebook Messenger, Instagram chats or WhatsApp), and Perplexity AI. Upcoming services include X’s Grok, slated as witty and irreverent, and Hume, billed as empathic AI. Recent Chinese developments include Baidu’s Ernie Bot (文心一言) and Alibaba’s Tongyi Tingwu. An experimental open-source tool built on a GPT base is AutoGPT, which can self-improve by writing its own code using GPT-4. Poe is a service which brings together multiple generative AI platforms, allowing users to compare results. DuckDuckGo offers private AI chat based on GPT and Claude, with a promise that chats are not saved or used to train AI.
Search engines > AI for search engines has been rolled out, effectively fusing conversational AI, generative AI and search functionality. The best-known examples are Microsoft’s Bing, linked to Copilot, and Google’s Gemini. Other AI-powered search options include the Opera browser’s Aria, while chatbots like ChatGPT and Perplexity AI can also serve as search engines. Services which search for and summarise scholarly research include Consensus, Elicit (which produces literature reviews), Scite.ai and Scholar AI.
Productivity software > AI is also making its way into productivity software, including Microsoft’s Copilot for Microsoft 365 (covering for example Word, PowerPoint, Outlook and Teams) and Google’s Gemini for Google Workspace (briefly named Duet before changing to its current name; this covers, for example, Gmail, Docs and Slides). Another similar project can be seen in the Notion suite of productivity tools incorporating Notion AI.
Specialised text-based services > There is a growing range of predominantly text-based generative AI software with different specialisations:
-
- composing (and editing) written documents: Jasper, Magic Write (Canva), Moonbeam, Rytr, Type
- editing (and composing) written documents: Grammarly, Lex, Trinka, Wordtune
- summarising/paraphrasing/creating notes: Mindgrasp, Quillbot
- asking questions of a PDF: ChatPDF, Sharly
- generating stories (including multimedia): DeepFiction, Depth Tale and Storly
- language learning (including audio): Duolingo Max, Kippy, LangAI, Speak, SpeakAI, Talkio, Univerbal, Vocalo, and the customised GPT Language Teacher
- generating (and interacting with) historical and other characters: Character AI, Hello History
- generating polls & quizzes: see lists on the Polling and Quizzes pages of this website
- generating courses/lesson plans/materials: see multimedia list below
Specialised multimedia services > There is also a growing range of multimedia generative AI software for creating and/or editing different artefact types:
-
- images: AI Image Generator (Canva), Copilot Designer (Microsoft), Imagine (Meta), DALL-E 2 [with DALL-E 2 now being phased out, and DALL-E 3 available through ChatGPT Plus or freely accessible through Microsoft’s Copilot Designer] (OpenAI), Firefly (Adobe), Freepik, Generative AI (Getty Images), Imagen AI (Google), Lexica, LogoFast (for logos), LogoMuse (customised GPT for logos), Midjourney, OpenArt, Photoleap, Scenery AI, Shakker (for remixing), Stable Diffusion (and the newer Stable Diffusion XL), Visual Electric; note that Lummi offers free stock images pre-created by generative AI
- mind maps: MyMap
- multimodal discussion boards: see list on the Discussion Boards page of this website
- QR codes with embedded images: see list on the QR Codes page of this website
- voice cloning: Audiobox (Meta), LOVO, PlayHT, Vall-e (X) (Microsoft) [research phase], Voice Engine (OpenAI) [research phrase]
- voice/video dubbing/translating: Dub AI, ElevenLabs, Translate Video
- music: Hydra, MusicGen (Meta), MusicLM (Google) [research phase], Musicfy, Stable Audio, Suno, Udio, Voicemod, Wavtool
- videos: see list on the Videos page of this website (note that some image software listed above also allows video editing)
- presentations: Decktopus, Gamma
- digital storytelling/comics: see list on the Digital Storytelling page of this website
- websites: see list on the Websites page of this website
- AI characters/avatars/chatbots: Artflow, Character.ai, Synthesia, Voiceflow
- multiple design formats: Magic Design (Canva)
- lesson designs/materials: Diffit, Education Copilot, EverLearns, Kaiden, Magic Design (Canva), MagicSchool, Twee (English focus)
As noted in the previous section, users can now build customised GPTs. More broadly, services like Promptly and Zapier allow users with little or no coding knowledge to build custom agents using generative AI, including but not limited to GPT.
Updates on AI developments are available from The Rundown, which also maintains a list of AI Super Tools. For an extensive, searchable catalogue of AI tools, classified by tool types (e.g., AI detection, generative art, generative video, research), see Future Tools, There’s an AI for That, Toolify or Top AI Tools.
Ongoing developments in generative AI
New iterations of LLMs and generative AI chatbots continue to be released on an almost daily basis, each promising improvements over the last. Work is currently underway on GPT-5, likely to be released in mid- to late 2024.
Digital/voice/smart assistants > Reports have emerged of first generation digital assistants being upgraded with generative AI capabilities: examples include Alexa, Bixby, Google Assistant and Siri.
Search engines > We are now seeing the arrival of AI-enabled smartphones which some commentators believe will remove the need for apps, with all user actions to be carried out through a phone digital assistant.
New form factors > In November 2023, Humane officially announced the upcoming release of its AI Pin, designed to allow hands-free interaction with AI via a lapel pin, with no need for a phone or other device. In the same month, Snapchat announced the incorporation of AI into its lenses. In February 2024, OpenAI announced a partnership with robotics company Figure, with a plan to integrate generative AI into robots; in March 2024, robotics company Covariant announced an AI robotics platform, RFM-1, for its warehouse robots; and in the same month, Nvidia announced its AI robotics platform, GR00T; while Sanctuary AI has released multiple generations of its Phoenix robot.
Prompting generative AI
The questions, commands and comments which users enter into conversations with generative AI are known as prompts, and it is vital for these to be as detailed and accurate as possible in order to obtain useful responses. Prompt literacy, indeed, might be seen as an extension of search literacy, especially given the ongoing fusion of generative AI with search services; and both overlap with AI literacy (see below). Prompt literacy is certainly a prerequisite for the emerging professional role which has come to be called prompt engineering, but in fact all users of generative AI need to develop skills in this area. Common prompting techniques include:
-
- zero-shot prompting (giving a simple question or command)
- one-shot prompting (giving a question-response example as a model)
- few-shot prompting (giving a few question-response examples as models)
- role prompting (giving the AI a specific role)
- chain of thought prompting (taking the AI through multiple steps involving reasoning)
Further details of these and other prompting techniques are available in From “Zero-Shot” To “Chain Of Thought” (Tristan Wolff/Medium, 2023), ChatGPT Guide: Use These Prompt Strategies to Maximize Your Results (Jonathan Kemper/The Decoder, 2024), and How to Write Better ChatGPT Prompts in 5 Steps (David Gewirtz/ZDNET, 2024).
A useful starting point is offered by The Rundown’s 6-Step Prompt Checklist, seen below, which helps users ensure their prompts are both specific and well contextualised:
For more information, see Master the 6-Step Prompt Formula (The Rundown, 2023). A copiable template for building good prompts can be seen at ChatGPT Prompting. Examples of effective prompts are available from a number of sites, though some such sites are now more oriented towards the generation of images, and some are moving towards freemium or fully paid models. Options include PromptHero, PromptPal, The Rundown’s 1000+ Copy-Paste ChatGPT Prompts. and, for educational prompts for both teachers and students, AI for Education’s GenAI Prompt Library for Educators. Many generative AI chatbots automatically suggest at least a few helpful prompts, and they can also be asked to provide more examples.
Generative AI in education
ChatGPT has many educational uses for students, including:
-
- explaining learning points (e.g., equations, scientific principles, grammar rules, historical events)
- providing examples of words, phrases or structures in use
- finding and summarising existing texts
- simplifying complex texts or information
- producing first drafts of essays and titles
- improving the grammar, vocabulary and style of texts
- co-generating stories (including in a choose-your-own-adventure style)
- offering timely feedback on texts
- creating self-study revision questions
- engaging in conversation (including role plays) in multiple languages
- taking on the role of a teacher or Socratic tutor (following specific instructions to this effect)
- taking on the role of a student (with students acting as teachers to improve its responses)
It has many additional educational uses for teachers, including:
-
- creating lesson plans
- creating teaching materials and handouts
- devising assessments and rubrics
- drafting model assignments
- generating multiple responses to a question (which students may critique)
- providing a first draft of feedback on student work (which students may critique)
- analysing student data to improve teaching and/or students’ learning
- drafting student feedback or reports based on teachers’ notes
- drafting meeting summaries based on notes
- drafting official documents based on notes
For suggestions on education-related AI tools, see Common Sense Education’s Classroom Tools that Use AI. For resources on the educational uses of ChatGPT and other generative AI, see Mike Sharples’ 2023 list in UNESCO’s ChatGPT and Artificial Intelligence in Higher Education: Quick Start Guide (p. 9), and Monash University’s AI in Education Learning Circle (also with a higher education focus). For ideas on how to prompt generative AI chatbots for educational uses, see 65+ Best ChatGPT Prompts for Teachers (Jill Baker/SplashLearn, 2024) and Unlocking Creative Lesson Plans with ChatGPT’s Prompts (Niall McNulty, 2023).
Note that a comprehensive understanding of generative AI requires digital literacies such as code literacy, and its subliteracies technological and AI literacy. AI literacy overlaps with relevant digital literacies such as text, multimodal and particularly search and prompt literacy (see above).
Generative AI in assessment
With the rise of generative AI, it is more important than ever for educators to be assessing students at the higher levels of Bloom’s Revised Taxonomy, ensuring that students go beyond remembering and understanding to applying, analysing, evaluating, and especially creating. Indeed, given the role of generative AI in today’s workplaces – where it can provide first drafts of standardised emails, weather or sports reports, legal opinions, coding, and multimedia advertising or architectural designs – and given its likely growing role in the future, it is appropriate for educators to assess students on their ability to engage in productive human-AI partnerships, where they improve on suggestions or drafts created by AI, or use AI to give feedback on, and help them hone, their own initial work.
For one approach to the continuum of possible AI uses in assessments, see Leon Furze’s 2023 The AI Assessment Scale: From No AI to Full AI, and for a consideration of assessments involving humans co-creating with AI, see Danny Liu and Adam Bridgeman’s 2023 ChatGPT is Old News: How Do We Assess in the Age of AI Writing Co-pilots?.
Certainly, in all cases where AI is used, it must be acknowledged. Because each user of ChatGPT (or other generative AI) may receive different answers at different times, the standard APA in-text reference follows this format: ‘ChatGPT, personal communication, [date]’. A similar referencing approach can be adopted with other AI chatbots. Note, however, that this will become rather more complex as AI is increasingly embedded into productivity tools.
Issues with generative AI
Pedagogical issues with generative AI include the need for prompt literacy (see above); the danger of hallucinations, where tools such as ChatGPT invent answers and even references (see Vectara’s regularly updated Hallucination Leaderboard for an overview of the percentage of hallucinations created by different generative AI tools); and issues around student plagiarism and creativity, with a strong implication that educators need to develop learning designs at the higher levels of the SAMR, T3 and PICRAT models and set assessments at the higher levels of Bloom’s Taxonomy. One of the co-creators of the well-known TPACK framework for teacher development, Punya Mishra, has suggested that TPACK can encompass generative AI.
Malevolent uses of AI include the creation of deepfakes to misrepresent people’s views or actions, including fake, sexually explicit materials for the purposes of exploitation or extortion; for more on this, see the Digital Safety & Wellness page of this website.
Broader issues include bias which is due, amongst other things, to the historical datasets on which generative AI is trained (see the work of Emily Bender, Timnit Gebru and colleagues); the black box nature of the technology, with even its programmers unable to say how it reaches its conclusions and generates its output; the unpaid labour of users whose freely shared online content is monetised by generative AI; and the extractive politics (see the work of Kate Crawford) and environmental racism (see again the work of Emily Bender, Timnit Gebru and colleagues) behind the development of generative AI, which exploits both human labour and planetary resources.
In July 2023, the Frontier Model Forum was launched by Anthropic, Google, Microsoft and Open AI to promote safe and responsible use of AI and to help leverage AI to address societal challenges. In December 2023, the AI Alliance was launched by IBM, Meta and other partners to also promote open, safe and responsible AI. In February 2024, the US National Institute of Standards and Technology announced the creation of the AI Safety Institute Consortium (AISIC), consisting of a large number of key technology companies and institutions. In March 2024, the European Union introduced the world’s first major Artificial Intelligence Act, designed to regulate AI development, and in the same month, the United Nations General Assembly adopted a resolution to promote safe development of AI. In the belief that strong, or general, AI may emerge in the not-too-distant future, OpenAI is working on a superalignment project with the aim of ensuring that the development and actions of AI are aligned with human values (though this inevitably raises the question of who has the power to define human values).
Last update: April 2024.
LATEST ON INSTAGRAM
Latest on Edublogs
- At the interface of AI and language learning October 20, 2023VicTESOL Symposium Melbourne, Australia 13 October, 2023 I was invited to be a member of a panel on Generative AI in EAL learning: Promises and challenges at the VicTESOL Symposium held at the Victorian Academy of Teaching and Leadership in North Melbourne. Hosted by Melissa Barnes (La Trobe University) and Katrina Tour (Monash University), the […]