what will bloomberg’s gpt ai built for finance mean for accountants?
by hitendra patil
rise of the aiccountants
bloomberg released a research paper detailing the development of bloomberggpt, a new large-scale generative artificial intelligence (ai) specifically trained on a wide range of financial data to support diverse natural language processing (nlp) tasks within the financial industry.
more: three ways a.i. will take shape in accounting | four ways to prepare for the ai era | seven ways ai could change accounting | who is better at accounting, ai or humans? | talent shortage: ai neither the cause nor the remedy | accounting profession to experience highest ai impact
exclusively for pro members. log in here or 2022世界杯足球排名 today.
according to bloomberg, “the team pulled from this extensive archive of financial data to create a comprehensive 363 billion token dataset consisting of english financial documents. this data was augmented with a 345 billion token public dataset to create a large training corpus with over 700 billion tokens. using a portion of this training corpus, the team trained a 50-billion parameter decoder-only causal language model.”
imagine how long it will take for a human being to grasp that volume of data!
what does it mean for the ai industry?
could this trigger an era of domain/industry-specific ai systems?
generative ai large language models (llms) like openai’s chatgpt and google’s bard are all-encompassing generic systems that professionals may find not too specific to their industries.
on the other hand, to create domain-industry-specific ai systems, the creators must have access to colossal volumes of industry-specific data representing nearly all situations that can occur in that domain/industry. and the systems would need generic ai capabilities as well to ensure the interactions can be used by professionals as well as non-professionals.
any entity with such industry-specific data will likely have deep pockets to create its own ai system to leverage that data. but the challenges of data biases cannot be ruled out, as one single entity with access to all types and sizes of businesses in an industry may be rare.
the demand for ai programmers, data scientists and research professionals will definitely skyrocket.
what could it mean for the accounting profession?
domain/industry-specific ai systems can transform how “users” interact with industry-specific data through artificially intelligent natural language queries. it is like a client asking a professional accountant a complex question, and the accountant uses expertise, experience and education to analyze the relevant financial data to answer the question.
in other words, data-driven answers may not need an interpreter as much as such interpreters are required in the non-ai software world.
now, relate that to the “advisory services” opportunities in the accounting profession. relate that to the day-to-day most common questions small business owners ask their accountants. relate that to the usual tax questions people ask their tax preparers.
no. it is not about the replacement of human expertise by ai. if you get paid to answer client questions rather than produce an outcome, then it will be a concern in the longer term. more likely than not, it is financially not too viable for any small to midsized accounting firm to be able to create ai tools to answer such client questions to reduce the time demand on their expert people. industry-specific ai tools are, therefore, likely to create more opportunities for experts to increase revenue from their expertise.
while writing my new book, “rise of the aiccountants,” i spent significant time researching, learning and understanding how ai systems are created, how they learn and work. three core ai concepts explained in my book stand out for the purpose of this post.
- supervised learning: at least for the foreseeable future, data quality (used for training the ai systems) and hence the accuracy of outputs of ai systems, and thus the reliability, may be a concern for accountants.
- unsupervised learning: similar is the case when ai systems continue to learn independently, based on user interactions (and likely from the data that gets fed into the ai system). accountants may not fully trust the industry-specific ai systems without adequate checks and balances and review/audit mechanisms for semi-supervised learning.
- explainability: the black box possibilities in ai systems mean ai models can arrive at conclusions and answers without explaining how such end result was reached. data confidentiality requirements may mean a reference to publicly available regulations may be given but not to the data from which ai systems learned. for client situations, accountants are used to providing professional explanations, complete with factual references. hence, the explainability issue in ai may become a real challenge, at least for some time.
will such an industry-specific ai future arrive?
ai technology will essentially be commercial endeavors. like in any business, return on investment considerations will drive the ai business creation decisions. the industry-specific regulatory environment will influence the speed at which such ai solutions emerge. talent shortage may also entice technology vendors to address the talent gap. it seems more likely than not that accounting and tax-specific ai solutions will emerge sooner than later. “when” is anybody’s guess.