Search-Bots : AI-Search-Centric LLM Archives
The website is dead, long live the AI-search-site.
Search-bots, LLMs hosted on and finetuned for a specific website- AI-search-site database knowledge domain, will offer persistent multimodal sharable research capacity. By constructing persistent, yet modifiable, custom displays for each user-defined search, LLMs will redefine humanities and science research design. This paradigm shift in online informational presentation will radically transform browser UX (user-experience) and revitalize archival research. Traditional websites with static, linked, and clickable pages will become legacy niches (fading away just like landlines, MOOs, horse-transportation, MySpace, printed encyclopedias, Flash etc…), replaced by spontaneously emergent configurations of sharable knowledge that allow for rapid, scalable, and persistent individualized and collective exploration.
Archival Digital Assistants ("ADAs") : the birth of a personalized data narrative
Search-Bots: imagine AI-first-chatbot-search at the core of archival research. Unlike traditional website search (“search” as a menu option that offers text-based results), the envisioned search-bot interface would be central and always-on, generating fully customizable, interactive, and integrated information specifically fit to to each user's ongoing search-conversational queries.
To clarify, the future of websites online archives lies not in human-designed webpages but in repositories of information accessed via Archival Digital Assistants ("ADAs") or search bots. These AI-systems will operate within certain constraints to create a personalized narrative of data for each user.
Case-study: ELMCIP (Electronic Literature Knowledge Base)
In the 2023 version of the online-knowledge-archive ELMCIP, a query like “Display the chronological evolution of digital literature works that use VR." produces nothing. The query “VR” produces a traditional list of articles scattered over many links, many pages.
Search-bot design would permit the user to request information on one page in a modality appropriate to their taste. "Display the chronological evolution of digital literature works using VR as a timeline" would dynamically construct an interactive timeline history. Refine: “Combine the timeline to contrast with AR. Highlight key works”. Continue: “Summarize the results by decade from 1990s to now with a paragraph for each decade, and provide a conclusion paragraph that outlines the different growth rates of AR and VR in digital literature.”
Open Exploration Informational Landscapes : research-journeys, search-path herstories
Logged in users could access their search-bot history, in essence creating journeys that others could follow like paths. And provide “custom instructions” to prepend and inform their preferences: “Always provide sources and links in a Chicago-style bibliography. Here’s a key writing sample in my style. Use my idiomatic style when I request my-style.” Each search query activates a search-bot to dynamically compile information into an interactive open-exploration custom-configuration layout. All this happens on a webpage search-site that is ‘growing’ above1 the central search-bot prompt-input-bar.
Toggle between different modes of interaction —text-based summary, list, mosaic, tiles, timelines, visualizations or a composite of all modes. Ask for your custom modes. A query for "N. Katherine Hayles" could manifest as a summary, key quotations, core ideas, an interactive timeline, and a network visualization illustrating the interconnected betweenness-centrality of her body of work.
Live linearly and calmly as a list of legacy links. Or, escape the linear infinite scroll, let the search-bot construct AR topology interface access to an ever-expanding informational landscape2. Spider out and search the web. Synthesize. Zoom in. Or out.
Conclusion: Search-bots will (displace websites and) become the dominant mode of digital research
The capability to implement search-bot archival-knowledge-systems is already within reach. As most acknowledge, large language models have advanced to a point where they can not only generate competent text but also capably write code, especially when templated and hosted on a single domain, enabling the dynamic construction of search-results in real-time. Fine-tuned on specific databases, aligned with domain-specific requirements and customized to fit interdisciplinary user needs, search-bots offer adaptable augmented research capacity.
Smaller search-bots (7B or 13B parameter models which run at feasible speeds on standard hardware) could be used for assembling info, coding, and rudimentary visualizations. Medium search-bots (40B or 70B parameters) could construct summaries (quotation compilations). Larger search-bots (GPT-4 API) could be used for insight-enhanced discursive-augmentation discourse-assist.
Summary: search-bots search-paths thru (re-)search-sites
In summary, search-bots have the potential to redefine website design and user-experience of the online digital landscape. By transforming static, text-based search results into dynamic, visually engaging search-paths, these systems offer a new paradigm for archival web research, UX design, pedagogical engagement, academic insights and agency-enhanced UX interactivity.
Cooper, Muriel, dir. 1994. Information Landscapes.
.