journal-chatbot.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "from langchain.llms import OpenAI\n",
    "from langchain.document_loaders import UnstructuredPDFLoader\n",
    "from langchain.vectorstores import Chroma\n",
    "from langchain.text_splitter import RecursiveCharacterTextSplitter\n",
    "\n",
    "from dotenv import load_dotenv\n",
    "_ = load_dotenv()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "from pathlib import Path\n",
    "from typing import Sequence\n",
    "from langchain.schema import Document\n",
    "\n",
    "article_path = \"data/On-algorithmic-fairness-in-medical-practice-2.pdf\"\n",
    "loader = UnstructuredPDFLoader(article_path)\n",
    "data: Sequence[Document] = loader.load()\n",
    "assert len(data) == 1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=0)\n",
    "texts: Sequence[Document] = text_splitter.split_documents(data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2023-07-09 23:19:23,413 INFO sqlalchemy.engine.Engine select pg_catalog.version()\n",
      "2023-07-09 23:19:23,421 INFO sqlalchemy.engine.Engine [raw sql] {}\n",
      "2023-07-09 23:19:23,429 INFO sqlalchemy.engine.Engine select current_schema()\n",
      "2023-07-09 23:19:23,438 INFO sqlalchemy.engine.Engine [raw sql] {}\n",
      "2023-07-09 23:19:23,456 INFO sqlalchemy.engine.Engine show standard_conforming_strings\n",
      "2023-07-09 23:19:23,456 INFO sqlalchemy.engine.Engine [raw sql] {}\n",
      "2023-07-09 23:19:23,478 INFO sqlalchemy.engine.Engine BEGIN (implicit)\n",
      "2023-07-09 23:19:23,486 INFO sqlalchemy.engine.Engine select relname from pg_class c join pg_namespace n on n.oid=c.relnamespace where pg_catalog.pg_table_is_visible(c.oid) and relname=%(name)s\n",
      "2023-07-09 23:19:23,492 INFO sqlalchemy.engine.Engine [generated in 0.00667s] {'name': 'vectorembedding'}\n",
      "2023-07-09 23:19:23,511 INFO sqlalchemy.engine.Engine \n",
      "CREATE TABLE vectorembedding (\n",
      "\tembedding VECTOR(1536), \n",
      "\tid SERIAL NOT NULL, \n",
      "\tdocument_name VARCHAR(100) NOT NULL, \n",
      "\tcontent VARCHAR NOT NULL, \n",
      "\tcreated_at TIMESTAMP WITHOUT TIME ZONE NOT NULL, \n",
      "\tPRIMARY KEY (id)\n",
      ")\n",
      "\n",
      "\n",
      "2023-07-09 23:19:23,511 INFO sqlalchemy.engine.Engine [no key 0.00340s] {}\n",
      "2023-07-09 23:19:23,595 INFO sqlalchemy.engine.Engine COMMIT\n"
     ]
    }
   ],
   "source": [
    "from langchain.embeddings import OpenAIEmbeddings\n",
    "from langchain.vectorstores.pgvector import PGVector\n",
    "from typing import Optional, Sequence\n",
    "from datetime import datetime\n",
    "from pgvector.sqlalchemy import Vector\n",
    "from sqlalchemy import Column\n",
    "\n",
    "# persist_directory = 'db'\n",
    "embedding = OpenAIEmbeddings()\n",
    "\n",
    "# from app.models.document import Document, VectorEmbedding\n",
    "# from app.models.user import User\n",
    "from sqlmodel import SQLModel, Session, Field, create_engine\n",
    "\n",
    "conn_string = \"postgresql://postgres:postgres@localhost:5432/postgres\"\n",
    "engine = create_engine(conn_string, echo=True)\n",
    "\n",
    "class PunkVectorEmbedding(SQLModel, table=True):\n",
    "    id: Optional[int] = Field(default=None, primary_key=True)\n",
    "    # document_id: int = Field(..., foreign_key=\"document.id\")\n",
    "    document_name: str = Field(..., max_length=100)\n",
    "    embedding: Sequence[float] = Field(..., sa_column=Column(Vector(1536)))\n",
    "    content: str = Field(...)\n",
    "    \n",
    "    created_at: datetime = Field(default_factory=datetime.now)\n",
    "\n",
    "try:\n",
    "    SQLModel.metadata.create_all(engine)\n",
    "except Exception as e:\n",
    "    print(e)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2023-07-09 23:19:27,621 INFO sqlalchemy.engine.Engine BEGIN (implicit)\n",
      "2023-07-09 23:19:27,846 INFO sqlalchemy.engine.Engine INSERT INTO vectorembedding (embedding, document_name, content, created_at) VALUES (%(embedding)s, %(document_name)s, %(content)s, %(created_at)s) RETURNING vectorembedding.id\n",
      "2023-07-09 23:19:27,846 INFO sqlalchemy.engine.Engine [generated in 0.21987s] ({'embedding': '[0.0065591825491602114,-0.004126159552925896,0.02907238956092918,-0.04080045673558116,0.01379974864657808,0.040552680222525575,-0.0187896828955745,0. ... (32561 characters truncated) ... 0.03380766575423498,-0.010874615258195313,-0.030614106568278544,-0.02108849439000587,-0.014866562843656967,0.009071355858552507,-0.02289175285832608]', 'document_name': 'On-algorithmic-fairness-in-medical-practice-2.pdf', 'content': 'Cambridge Quarterly of Healthcare Ethics (2022), 31: 1, 83–94 doi:10.1017/S0963180121000839\\n\\nB I O E T H I C S A N D I N F O R M A T I O N T E C H  ... (503 characters truncated) ... 2Leverhulme Centre for the Future of Intelligence, University of Cambridge, Cambridge, UK *Corresponding author. Email: thomas.grote@uni-tuebingen.de', 'created_at': datetime.datetime(2023, 7, 9, 23, 19, 27, 212710)}, {'embedding': '[0.00013987326162841546,0.004377609277519737,0.027628784138226413,-0.03805909071256678,0.005049000677222414,0.01649317763933492,-0.02232546812765669, ... (32558 characters truncated) ... 02299007856744776,-0.010098001354444828,-0.032443815155804155,-0.010532033080208333,-0.004862503324894617,0.005238889440828621,-0.010911809676098137]', 'document_name': 'On-algorithmic-fairness-in-medical-practice-2.pdf', 'content': 'Abstract The application of machine-learning technologies to medical practice promises to enhance the capabilities of healthcare professionals in the ... (580 characters truncated) ... ness; machine learning; algorithmic bias; discrimination; medical practice\\n\\nIntroduction\\n\\ns s e r P y t i s r e v i n U e g d i r b m a C y b e n', 'created_at': datetime.datetime(2023, 7, 9, 23, 19, 27, 229584)}, {'embedding': '[-0.01368028487890777,0.022653805151688845,0.031547322952454424,-0.038534134727958226,-0.0077868290450781065,0.021440446433314046,-0.0065101357348496 ... (32594 characters truncated) ... .010273546938855455,-0.019973749347524588,-0.03906748081610453,0.003706743950481392,-0.01079355835034327,-0.016427008478429025,0.0010458551141992094]', 'document_name': 'On-algorithmic-fairness-in-medical-practice-2.pdf', 'content': 'Recent developments in machine learning have seen the application of data-driven technologies to the assessment, diagnosis, and treatment, of medical ... (600 characters truncated) ...  a mental-health problem, such as predicting a post-traumatic stress course for patients suffering from a traumatic event.3\\n\\ni l\\n\\nn o d e h s i l', 'created_at': datetime.datetime(2023, 7, 9, 23, 19, 27, 245584)}, {'embedding': '[-0.005966152407935758,0.01590524208743382,0.0216117031170002,-0.0396349459217173,0.0046946773275065605,0.027399107252513287,-0.002556440600905966,-0 ... (32611 characters truncated) ... 019615117058636986,-0.021895003987813685,-0.03521007972346857,-0.014259407935968757,-0.012761966995212514,-0.006094311704803011,-0.00454628225415263]', 'document_name': 'On-algorithmic-fairness-in-medical-practice-2.pdf', 'content': 'These algorithms thus lay the ground for advanced decision support systems that enable physicians to minimize diagnostic errors. Finally, there are h ... (545 characters truncated) ...  before\\n\\n.\\n\\n© The Author(s), 2022. Published by Cambridge University Press.\\n\\n. i\\n\\no d / / : s p t t h\\n\\n84\\n\\nThomas Grote and Geoff Keeling', 'created_at': datetime.datetime(2023, 7, 9, 23, 19, 27, 263526)}, {'embedding': '[-0.016937551105635964,-0.0056514072602183215,0.03358169460564698,-0.03475532130285625,-0.0030841014212102612,0.007828616044152538,-0.004531128120053 ... (32602 characters truncated) ... -0.01711092798847884,-0.02173874648930416,-0.03400846761142394,-0.01208300863058433,-0.011042750127494956,-0.009035583160077187,0.007728591062715898]', 'document_name': 'On-algorithmic-fairness-in-medical-practice-2.pdf', 'content': 'machine-learning systems become an integral part of medical practice. Although chances are high that the incorporation of machine learning will impro ... (272 characters truncated) ... r people of non-European ancestry.6 The bottom line is that due to these biases, they are at risk of receiving unfair treatment in clinical settings.', 'created_at': datetime.datetime(2023, 7, 9, 23, 19, 27, 271526)}, {'embedding': '[-0.0030095215705011817,-0.020269085853837424,0.038495345328485794,-0.04788704111401605,0.00801212571605785,0.019990517942986095,-0.02135682411561004 ... (32584 characters truncated) ... 0.013915099049425932,-0.0229221073674135,-0.03788515019409877,-0.01919461148891327,-0.009749854528053391,-0.005176708857008091,-0.027246532620807028]', 'document_name': 'On-algorithmic-fairness-in-medical-practice-2.pdf', 'content': 'The implementation of machine learning thus threatens to exacerbate existing inequalities in healthcare, prompting urgent reflection on the means to  ... (656 characters truncated) ... ntrast, we address the problem of algorithmic fairness at the population level while making clear why our considerations matter for medical practice.', 'created_at': datetime.datetime(2023, 7, 9, 23, 19, 27, 278540)}, {'embedding': '[-0.0012626927743001298,0.01078672368466941,0.026754730290959147,-0.02223905245847326,0.021377681495354316,0.014225685484950103,-0.020020368207143386 ... (32619 characters truncated) ... .01678369805833271,-0.010734519327430529,-0.03829189137810683,-0.008626768636741325,0.004434107010900901,-0.008881265576772825,-0.013240328474896859]', 'document_name': 'On-algorithmic-fairness-in-medical-practice-2.pdf', 'content': 'With that in mind, our paper has three objectives:\\n\\n1) We try to provide an account of the ways that algorithmic discrimination manifests in medica ... (54 characters truncated) ... mechanisms of algorithmic bias. 3) We try to identify the appropriate normative standards for fair algorithmic decisionmaking in\\n\\nmedical practice.', 'created_at': datetime.datetime(2023, 7, 9, 23, 19, 27, 283885)}, {'embedding': '[-0.007332464993920447,-0.0016565009375787304,0.024374586956683843,-0.03857077578316789,0.005608164123559452,0.009816798178253539,-0.0254727820248359 ... (32556 characters truncated) ... 20316620868007652,-0.015294383894636178,-0.025526353560863826,-0.016901500173149876,0.0006708870502982039,-0.011129275665226845,-0.03500833662386247]', 'document_name': 'On-algorithmic-fairness-in-medical-practice-2.pdf', 'content': 'In addition, one upshot of the paper is that it ties together certain themes in the emerging literature on “fairness in machine learning” with the ph ... (700 characters truncated) ... ider array of normative criteria needs to be taken into account. By discussing the constraints of formal accounts of algorithmic fairness, we hope to', 'created_at': datetime.datetime(2023, 7, 9, 23, 19, 27, 288074)}  ... displaying 10 of 70 total bound parameter sets ...  {'embedding': '[-0.010418722014222241,-0.025409595085289245,0.023780438154716067,-0.014241986372433083,-0.01180481959444506,0.01757913016729813,-0.01739519396572869 ... (32585 characters truncated) ... 018446262232171357,-0.024949752718720578,-0.003296086906368399,-0.00617503072196335,-0.010569813263787881,0.027222690080076935,-0.025475286851941913]', 'document_name': 'On-algorithmic-fairness-in-medical-practice-2.pdf', 'content': '40. Hooker B. Fairness. Ethical Theory and Moral Practice 2005;8:329–30. 41. Cf. Saunders B. Fairness between competing claims. Res Publica 2010;16:4 ... (165 characters truncated) ... 0; Broome J. Kamm on fairness. Philosophy and Phenomenological Research 1998;58:955–61; see note 35, Broome 2004, at 37–40; see note 36, Hooker 2005.', 'created_at': datetime.datetime(2023, 7, 9, 23, 19, 27, 613855)}, {'embedding': '[0.023317443689252995,-0.011294824333943106,0.01685126212324381,-0.027642227935630287,0.0221417756682659,0.019118624538926123,-0.02153994388030449,0. ... (32557 characters truncated) ... 759342954542238,-0.00020097474309225426,-0.0064486871381419205,-0.032330909692690965,-0.015311695660250802,0.012729421435764965,-0.01967846671977403]', 'document_name': 'On-algorithmic-fairness-in-medical-practice-2.pdf', 'content': '43. See note 38, Broome 1998, at 959. 44. See note 35, Broome 2004. 45. Note that this is a simplification. Although claims to medical resources are  ... (705 characters truncated) ... ic Fairness in Medical Practice. Cambridge Quarterly of Healthcare Ethics 31: 83–94, doi:10.1017/S0963180121000839\\n\\n.\\n\\n. i\\n\\no d / / : s p t t h', 'created_at': datetime.datetime(2023, 7, 9, 23, 19, 27, 613855)})\n",
      "2023-07-09 23:19:28,091 INFO sqlalchemy.engine.Engine COMMIT\n"
     ]
    }
   ],
   "source": [
    "with Session(engine) as session:\n",
    "    chunks_to_embed = [text.page_content for text in texts]\n",
    "    vectors = embedding.embed_documents(chunks_to_embed)\n",
    "    db_vectors = [\n",
    "        PunkVectorEmbedding(\n",
    "            document_name=Path(article_path).name, embedding=vector, content=chunk\n",
    "        )\n",
    "        for vector, chunk in zip(vectors, chunks_to_embed)\n",
    "    ]\n",
    "    for db_vector in db_vectors:\n",
    "        session.add(db_vector)\n",
    "    session.commit()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "from langchain.chains import VectorDBQA, RetrievalQA, ConversationalRetrievalChain\n",
    "from langchain.llms import OpenAI\n",
    "from langchain.chat_models import ChatOpenAI\n",
    "from langchain.memory import ConversationBufferMemory\n",
    "\n",
    "# qa = VectorDBQA.from_chain_type(\n",
    "#     llm=OpenAI(model=\"gpt-3.5-turbo\"),\n",
    "#     chain_type=\"stuff\",\n",
    "#     vectorstore=PGVector(\n",
    "#         connection_string=conn_string,\n",
    "#         embedding_function=embedding,\n",
    "#         collection_name=\"public\",\n",
    "#     ),\n",
    "# )\n",
    "# qa = RetrievalQA(\n",
    "#     llm=OpenAI(model=\"gpt-3.5-turbo\"),\n",
    "# )\n",
    "# vectorstore = PGVector(\n",
    "#     connection_string=conn_string,\n",
    "#     embedding_function=embedding,\n",
    "#     collection_name=\"public\",\n",
    "# )\n",
    "# memory = ConversationBufferMemory(memory_key=\"chat_history\", return_messages=True)\n",
    "# qa = ConversationalRetrievalChain.from_llm(\n",
    "#     ChatOpenAI(temperature=0, model=\"gpt-4\"), \n",
    "#     vectorstore.as_retriever(),\n",
    "#     condense_question_llm=ChatOpenAI(temperature=0, model='gpt-3.5-turbo'),\n",
    "#     return_source_documents=True,\n",
    "# )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "# vs_retriever = vectorstore.as_retriever()\n",
    "# vs_retriever.get_relevant_documents(\"What is the meaning of life?\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "# qa.run(\"Can you explain the problem with fairness?\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "# result = qa(\n",
    "#     {\n",
    "#         \"question\": \"Please provide a two-paragraph summary of the entire paper.\",\n",
    "#         \"chat_history\": [],\n",
    "#     }\n",
    "# )\n",
    "# result"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "# TODO:\n",
    "# 1. Add a function that makes the question into an embedding\n",
    "# 2. Extract the relevant documents via direct SQL query (compare on vector distance, but get the actual text)\n",
    "# 3. Combine the document(s) and the question into a single GPT-4 query"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "from langchain.embeddings.base import Embeddings\n",
    "from sqlmodel import select\n",
    "\n",
    "\n",
    "def get_embedding_from_query(query: str, embedder: Embeddings) -> Sequence[float]:\n",
    "    embeddings = embedder.embed_documents([query])\n",
    "    return embeddings[0]\n",
    "\n",
    "\n",
    "def get_k_similar_chunks(\n",
    "    query_embedding: Sequence[float], session: Session, k: int = 3\n",
    ") -> Sequence[str]:\n",
    "    # query = session.query(\n",
    "    #     f\"SELECT content FROM vectorembeddings ORDER BY embedding <=> %s LIMIT {k}\",\n",
    "    #     query_embedding,\n",
    "    # )\n",
    "    query = (\n",
    "        select(PunkVectorEmbedding.content)\n",
    "        .order_by(PunkVectorEmbedding.embedding.l2_distance(query_embedding))\n",
    "        .limit(k)\n",
    "    )\n",
    "    results = session.exec(query)\n",
    "\n",
    "    return results\n",
    "\n",
    "\n",
    "TEMPLATE = \"\"\"\n",
    "You are a good chatbot that answers questions regarding published journal papers. You are academic and precise.\n",
    "\n",
    "The relevant chunks of the paper are:\n",
    "\n",
    "- {chunks}.\n",
    "\n",
    "Please respond to the following question or request: {query}\n",
    "\"\"\"\n",
    "\n",
    "\n",
    "def create_query(\n",
    "    query: str, session: Session, embedder: Embeddings, template: str = TEMPLATE\n",
    ") -> str:\n",
    "    query_embedding = get_embedding_from_query(query, embedder=embedding)\n",
    "    relevant_chunks = get_k_similar_chunks(query_embedding, session)\n",
    "\n",
    "    connected_chunks = \"\\n- \".join(relevant_chunks)\n",
    "\n",
    "    final_query = template.format(chunks=connected_chunks, query=query)\n",
    "\n",
    "    return final_query"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2023-07-09 23:19:28,636 INFO sqlalchemy.engine.Engine BEGIN (implicit)\n",
      "2023-07-09 23:19:28,654 INFO sqlalchemy.engine.Engine SELECT vectorembedding.content \n",
      "FROM vectorembedding ORDER BY vectorembedding.embedding <-> %(embedding_1)s \n",
      " LIMIT %(param_1)s\n",
      "2023-07-09 23:19:28,660 INFO sqlalchemy.engine.Engine [generated in 0.01091s] {'embedding_1': '[-0.025741690304702535,0.018027016567325336,0.016134245029648445,-0.001245802595123262,-0.0037365918303938436,0.013060122492383468,-0.028561267485762 ... (32628 characters truncated) ... ,-0.03305170625336563,0.002491605190246524,-0.044303904418377,0.006761762657037608,0.0007599640383914042,-0.011950567148775315,-0.010390662400445384]', 'param_1': 3}\n",
      "2023-07-09 23:19:28,719 INFO sqlalchemy.engine.Engine ROLLBACK\n"
     ]
    }
   ],
   "source": [
    "# query_embedding = get_embedding_from_query(\n",
    "#     \"Please provide a two-paragraph summary of the entire paper\", embedder=embedding\n",
    "# )\n",
    "# chunks = get_k_similar_chunks(query_embedding, session)\n",
    "initial_query = \"Please provide a two-paragraph summary of the entire paper\"\n",
    "with Session(engine) as session:\n",
    "    final_query = create_query(\n",
    "        query=initial_query,\n",
    "        session=session,\n",
    "        embedder=embedding\n",
    "    )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "You are a good chatbot that answers questions regarding published journal papers. You are academic and precise.\n",
      "\n",
      "The relevant chunks of the paper are:\n",
      "\n",
      "- In addition, one upshot of the paper is that it ties together certain themes in the emerging literature on “fairness in machine learning” with the philosophical debate on justice in healthcare. The remainder of the paper will be structured as follows: In the section “Mechanisms of Algorithmic Bias in Medical Practice,” we give an outline of the different mechanisms of algorithmic bias in medical practice, whereby we distinguish between formal, substantive, and normative notions of algorithmic bias. Thereafter, the section “From Algorithmic Bias to Fairness” will discuss to what extent different standards of fairness help to counteract the threat of algorithmic discrimination. Here, we argue that fairness cannot be restored merely by mitigating the differences in the algorithm’s predictive accuracy for different demographics. Instead, a wider array of normative criteria needs to be taken into account. By discussing the constraints of formal accounts of algorithmic fairness, we hope to\n",
      "- With that in mind, our paper has three objectives:\n",
      "\n",
      "1) We try to provide an account of the ways that algorithmic discrimination manifests in medical\n",
      "\n",
      "practice.\n",
      "\n",
      "2) We try to examine the underlying mechanisms of algorithmic bias. 3) We try to identify the appropriate normative standards for fair algorithmic decisionmaking in\n",
      "\n",
      "medical practice.\n",
      "- There are two lessons to be drawn from the study. The first one is simple: developers should be wary of the larger social context when choosing the target category for an algorithm. By contrast, the second one is that there may be profound problems in the way that machine-learning algorithms operate, namely by making predictions in virtue of exploiting correlations in the data while being unable to capture potentially discriminatory factors. We will address this issue in the subsequent chapter. In sum, the substantive notion helps us to understand the various pragmatic and normative constraints which give rise to algorithmic bias. It does not, however, grasp the whole morally normative dialectic of algorithmic bias..\n",
      "\n",
      "Please respond to the following question or request: Please provide a two-paragraph summary of the entire paper\n",
      "\n"
     ]
    }
   ],
   "source": [
    "print(final_query)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "# TODO: Pass the final query to GPT-4 via either Langchain or direct OpenAI API\n",
    "from langchain.llms import OpenAI\n",
    "from langchain.chat_models import ChatOpenAI\n",
    "from langchain.schema import (\n",
    "    AIMessage,\n",
    "    HumanMessage,\n",
    "    SystemMessage\n",
    ")\n",
    "\n",
    "\n",
    "chat = ChatOpenAI(model=\"gpt-4\", temperature=0)\n",
    "# TODO: Split up query into system message and human message?\n",
    "response = chat([\n",
    "    HumanMessage(content=final_query)\n",
    "])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "This paper is an in-depth exploration of algorithmic bias in medical practice, linking themes from the burgeoning field of fairness in machine learning to philosophical discussions on justice in healthcare. It aims to outline the different mechanisms of algorithmic bias, distinguishing between formal, substantive, and normative notions. The paper also scrutinizes the extent to which different fairness standards can combat the threat of algorithmic discrimination, arguing that fairness restoration requires more than just mitigating predictive accuracy differences across various demographics. The constraints of formal accounts of algorithmic fairness are also discussed, suggesting a need for consideration of a wider array of normative criteria.\n",
      "\n",
      "The paper sets three objectives: to provide an understanding of how algorithmic discrimination manifests in medical practice, to examine the underlying mechanisms of algorithmic bias, and to identify the suitable normative standards for fair algorithmic decision-making in medical practice. Two significant takeaways from the study are that developers should consider the broader social context when defining the target category for an algorithm, and that there may be deep-rooted issues in the functioning of machine-learning algorithms. These issues arise due to the algorithm's reliance on exploiting data correlations without being able to capture potentially discriminatory factors, a topic that is addressed in subsequent sections of the paper.\n"
     ]
    }
   ],
   "source": [
    "print(response.content)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "venv",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.10.11"
  },
  "orig_nbformat": 4
 },
 "nbformat": 4,
 "nbformat_minor": 2
}